Remove outdated items: Gen3R (06, 16), P4 streaming (01, 02), Lyra-era fixers (14, 15), superseded arms (08, 09, 10)

Superseded by the current direction (unified recon + gen + refine with a 3D latent-root memory). Everything removed here is still in the history: restore from revision 7e37680de15f60ede96e125372b963fe7304eb17.

01_p4_mixed_v1.zip

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:9a67a6036956c6d32cafe001e05e803f7c1f6895f5056052058cb58538eb8460
-size 5627794778

02_p4_control_v1.zip

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:a68adbdd8bc8a097e91167d6361084d1cb743dedbafbf36d07fc38cf2a42b413
-size 5627794790

06_gen3r_outputs.zip

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:fe7959a3bd5770cbd8075d34a97185c3fdf5aa1c7395f95b37403d1dcd622f42
-size 6834652684

08_clean_pixel_c3r/README.md

@@ -1,27 +0,0 @@
-# Clean pixel arm: C3 + realism, EMA at step 60000
-
-`ema_060000.pt` - the pixel-root arm of the clean line (outputs/clean/c3r_main/ema_060000.pt), trained from scratch as the
-matched control of the voxel arm in `07_clean_voxel_f3r/`.
-
-- Matched to the voxel arm: same 3D-Belief trunk, same decoder warm start (free_completion_20260927 ray/step002000), same
-  data (official RE10K train only, online VGGT-Omega teacher vggt_omega_1b_512, fresh 11-frame clip per visit), same
-  realism losses (--drift-global --drift-geo --ot-p 0.5 --ot-mode lateral --ot-frac 0.3), seed 1, 8 H100 x 1 clip per
-  step, constant lr (wsd, no cooldown). The only difference: no voxel head / fuser / voxel stream (pixel roots, coverage
-  memory rule) - the five voxel flags of the voxel arm are absent.
-- Stopped at 60k by decision (2026-10-28): still creeping (visible LPIPS W3 .241 -> .232 over 45-55k, .237 at 60k),
-  far behind the voxel arm on observed metrics at matched steps. Not a converged plateau.
-- Full command: command.json. Code: branch clean-re10k-online-teacher at f1252bf plus uncommitted changes to
-  flc/unified/train_v4.py (`--accum` gradient accumulation, unused here at accum 1; redraw of clips with truncated PNGs):
-  train_v4_vs_f1252bf.patch.
-- Held-out (40 clean RE10K episodes, unified 4, blind, T=1.0, coverage memory rule, no update head): eval_summary_40ep.json.
-
-| Step 60k, no update head | Visible LPIPS W1 / W3 | Revisit-A PSNR W3 | Reveal LPIPS W2 | Whole KID | Patch KID occluded / out-of-view |
-|---|---|---|---|---|---|
-| pixel arm (this) | 0.263 / 0.237 | 23.59 dB | 0.303 | 0.0354 | 0.058 / 0.026 |
-| voxel arm, same step (f3r_main 60k) | 0.214 / 0.206 | 25.10 dB | 0.207 | 0.0014 | 0.138 / 0.039 |
-
-Pixel leads only on hidden-region patch KID. Update heads for both arms at 60k (upd.sbatch recipe) are pending.
-
-- Load as an evaluate_stream_amodal.py --snapshot (dict with 'ema', 'step', 'args'), with --mem-rule coverage.
-
-sha256 ema_060000.pt: e15fc5a7149802bbe7677ee6e9369b44e7f55cc54a28a4d2dc71faf366e1d820

08_clean_pixel_c3r/command.json

@@ -1,171 +0,0 @@
-{
- "argv": [
-  "flc/unified/train_v4.py",
-  "--out",
-  "outputs/clean/c3r_main",
-  "--accum",
-  "1",
-  "--decoder-from",
-  "outputs/clean/free_completion_20260927/checkpoints/ray/step002000",
-  "--online-scenes",
-  "data/re10k_scenes/train",
-  "--lr-schedule",
-  "wsd",
-  "--cooldown-from",
-  "-1",
-  "--steps",
-  "500000",
-  "--warmup",
-  "1000",
-  "--save-every",
-  "500",
-  "--keep",
-  "5000,10000,15000,20000,25000,30000,35000,40000,45000,50000,55000,60000,65000,70000,75000,80000,85000,90000,95000,100000,105000,110000,115000,120000,125000,130000,135000,140000,145000,150000,155000,160000,165000,170000,175000,180000,185000,190000,195000,200000,205000,210000,215000,220000,225000,230000,235000,240000,245000,250000,255000,260000,265000,270000,275000,280000,285000,290000,295000,300000,305000,310000,315000,320000,325000,330000,335000,340000,345000,350000,355000,360000,365000,370000,375000,380000,385000,390000,395000,400000,405000,410000,415000,420000,425000,430000,435000,440000,445000,450000,455000,460000,465000,470000,475000,480000,485000,490000,495000,500000",
-  "--viz-every",
-  "5000",
-  "--seed",
-  "1",
-  "--depth-w",
-  "1",
-  "--adv-w",
-  "0.1",
-  "--gate",
-  "--workers",
-  "6",
-  "--held",
-  "3",
-  "--dis-p",
-  "0.5",
-  "--layers",
-  "1",
-  "--p-uncond",
-  "0.1",
-  "--mixed-p",
-  "0.4",
-  "--drift-p",
-  "1",
-  "--drift-n",
-  "8",
-  "--drift-no-held",
-  "--recon-p",
-  "0.25",
-  "--drift-global",
-  "--drift-geo",
-  "--ot-p",
-  "0.5",
-  "--ot-mode",
-  "lateral",
-  "--ot-frac",
-  "0.3"
- ],
- "args": {
-  "out": "outputs/clean/c3r_main",
-  "steps": "500000",
-  "cache": "/data/team-jhu/yyin34/free-latent-completion/data/local_re10k/cache",
-  "stream_p": "0.0",
-  "fuse": "False",
-  "fuse_tol": "0.05",
-  "revisit_w": "1.0",
-  "p_uncond": "0.1",
-  "recon_p": "0.25",
-  "mixed_p": "0.4",
-  "vggt_cache": "",
-  "online_scenes": "data/re10k_scenes/train",
-  "online_extra": "[]",
-  "online_weight": "1.0",
-  "online_span": "(45, 150)",
-  "teacher_weights": "/data/team-jhu/yyin34/free-latent-completion/checkpoints/vggt_omega_1b_512.pt",
-  "lr_schedule": "wsd",
-  "cooldown_from": "-1",
-  "exclude": "/data/team-jhu/yyin34/free-latent-completion/configs/splits/local_re10k/val.txt,/data/team-jhu/yyin34/free-latent-completion/configs/splits/local_re10k/test.txt",
-  "depth_w": "1.0",
-  "ray_disp": "False",
-  "gate": "True",
-  "layers": "1",
-  "held": "3",
-  "dis_p": "0.5",
-  "smooth_w": "0.001",
-  "visible_w": "0.0",
-  "chunks": "",
-  "p_chunks": "0.7",
-  "train_list": "/data/team-jhu/yyin34/free-latent-completion/configs/splits/local_re10k/train.txt",
-  "decoder_from": "outputs/clean/free_completion_20260927/checkpoints/ray/step002000",
-  "lr_trunk": "2e-05",
-  "lr_head": "0.0002",
-  "warmup": "1000",
-  "repa_w": "1.0",
-  "self_w": "0.0",
-  "self_views": "2",
-  "self_alpha_w": "0.1",
-  "self_valid": "0.2",
-  "lpips_w": "1.0",
-  "identity_w": "0.8",
-  "held_w": "1.0",
-  "adv_w": "0.1",
-  "ot_p": "0.5",
-  "ot_n": "2",
-  "ot_deg": "40.0",
-  "ot_mode": "lateral",
-  "ot_frac": "0.3",
-  "ot_adv_w": "0.1",
-  "ot_geo_w": "1.0",
-  "orient_w": "0.0",
-  "adv_start": "0",
-  "adv_min_t": "0.3",
-  "adv_mask": "False",
-  "ht_c": "0.0",
-  "scale_bg": "0.5",
-  "ht_group": "False",
-  "drift_p": "1.0",
-  "drift_n": "8",
-  "drift_min_t": "0.4",
-  "drift_grad": "2",
-  "drift_w": "1.0",
-  "drift_anchor_w": "1.0",
-  "drift_reg": "none",
-  "drift_reg_w": "1.0",
-  "train_merge": "none",
-  "train_merge_tol": "0.05",
-  "voxel_head": "",
-  "voxel_size": "0.1",
-  "voxel_hidden": "0",
-  "voxel_stream": "False",
-  "voxel_anchor": "False",
-  "voxel_select": "False",
-  "voxel_frame": "member",
-  "voxel_occ_w": "0.1",
-  "voxel_sdf_w": "1.0",
-  "voxel_normal_w": "1.0",
-  "voxel_fuser": "False",
-  "voxel_gen_realism": "False",
-  "drift_geo": "True",
-  "drift_global": "True",
-  "corr": "False",
-  "corr_p": "1.0",
-  "corr_w": "1.0",
-  "corr_k": "4",
-  "drift_no_held": "True",
-  "ht_split_oov": "False",
-  "ht_hidden_only": "False",
-  "lpips_group": "False",
-  "adv_hidden_w": "0.0",
-  "onpolicy_p": "0.0",
-  "onpolicy_steps": "10",
-  "onpolicy_temp": "1.0",
-  "adv_mask_onpol": "False",
-  "osft_p": "0.0",
-  "osft_n": "2",
-  "osft_tau": "0.5,0.8",
-  "osft_steps": "10",
-  "dual_2d": "0.0",
-  "init_from": "",
-  "save_every": "500",
-  "viz_every": "5000",
-  "keep": "5000,10000,15000,20000,25000,30000,35000,40000,45000,50000,55000,60000,65000,70000,75000,80000,85000,90000,95000,100000,105000,110000,115000,120000,125000,130000,135000,140000,145000,150000,155000,160000,165000,170000,175000,180000,185000,190000,195000,200000,205000,210000,215000,220000,225000,230000,235000,240000,245000,250000,255000,260000,265000,270000,275000,280000,285000,290000,295000,300000,305000,310000,315000,320000,325000,330000,335000,340000,345000,350000,355000,360000,365000,370000,375000,380000,385000,390000,395000,400000,405000,410000,415000,420000,425000,430000,435000,440000,445000,450000,455000,460000,465000,470000,475000,480000,485000,490000,495000,500000",
-  "seed": "1",
-  "mix": "",
-  "workers": "6",
-  "accum": "1",
-  "world_size": "8"
- }
-}

08_clean_pixel_c3r/ema_060000.pt

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:e15fc5a7149802bbe7677ee6e9369b44e7f55cc54a28a4d2dc71faf366e1d820
-size 1875933734

08_clean_pixel_c3r/train_v4_vs_f1252bf.patch

@@ -1,127 +0,0 @@
-diff --git a/flc/unified/train_v4.py b/flc/unified/train_v4.py
-index 217af03..3c6b1c5 100644
---- a/flc/unified/train_v4.py
-+++ b/flc/unified/train_v4.py
-@@ -99,8 +99,16 @@ class OnlineClips(torch.utils.data.Dataset):
-         s = torch.load(self.files[i], weights_only=False)
-         self.draws += 1
-         rng = random.Random(hash((torch.initial_seed(), i, self.draws)))
--        ids = pick_frames(s['n'], 11, rng, self.spans[i])
--        ims = [cv2.imdecode(np.frombuffer(s['images'][k].numpy().tobytes(), np.uint8), cv2.IMREAD_COLOR)[:, :, ::-1] for k in ids]
-+        dec = lambda k: cv2.imdecode(np.frombuffer(s['images'][k].numpy().tobytes(), np.uint8), cv2.IMREAD_COLOR)
-+        for _ in range(4):                         # a few source frames are truncated PNGs: redraw the clip around them
-+            ids = pick_frames(s['n'], 11, rng, self.spans[i])
-+            ims = [dec(k) for k in ids]
-+            if all(m is not None for m in ims):
-+                break
-+        else:
-+            print(f'OnlineClips: undecodable frames in {self.files[i]}, using another scene', flush=True)
-+            return self[rng.randrange(len(self.files))]
-+        ims = [m[:, :, ::-1] for m in ims]
-         sq = lambda size: torch.from_numpy(np.stack([cv2.resize(m, (size, size), interpolation=cv2.INTER_AREA) for m in ims])).permute(0, 3, 1, 2).contiguous()
-         return dict(f512=sq(512), img=sq(SHAPE[0]), frame=torch.tensor(ids), scene=s['key'])
- 
-@@ -746,12 +754,14 @@ def main():
-     ap.add_argument('--seed', type=int, default=0)
-     ap.add_argument('--mix', default='', help='sampling weight per --vggt-cache dir, e.g. 0.7,0.3 (default: uniform over clips)')
-     ap.add_argument('--workers', type=int, default=6)
-+    ap.add_argument('--accum', type=int, default=1, help='clips per GPU per step (gradient accumulation): N GPUs x --accum K '
-+                    'is the N*K-GPU batch (micro-step m of rank r samples the tasks of rank r + m*N; clips from the rank\'s loader)')
-     args = ap.parse_args()
-     # --ht-hidden-only leaves the visible scales untrained (visible patches use MSE), so --lpips-group's reference stays
-     # at its 0.1 init and up-weights LPIPS on hidden pixels 1.3-5x instead of lowering it (found in the A2c+ run, 2026-10-16)
-     assert not (args.ht_hidden_only and args.lpips_group), '--lpips-group needs trained visible scales; drop --ht-hidden-only'
-     assert args.drift_reg != 'ht' or (args.ht_c > 0 and args.ht_group), '--drift-reg ht needs --ht-c and --ht-group'
--    # Multi-GPU (torchrun): one clip per GPU per step, gradients averaged across GPUs; rank 0 logs, saves and keeps
-+    # Multi-GPU (torchrun): one clip per GPU per step (--accum K: K clips), gradients averaged across GPUs; rank 0 logs, saves and keeps
-     # the EMA. Every rank starts from rank 0's weights; data, task and noise streams differ per rank.
-     world, rank = int(os.environ.get('WORLD_SIZE', 1)), int(os.environ.get('RANK', 0))
-     if world > 1:
-@@ -901,7 +911,9 @@ def main():
-         (args.out/'command.json').write_text(json.dumps(dict(argv=sys.argv, args={k: str(v) for k, v in vars(args).items()}), indent=1))
-     keep = {int(x) for x in args.keep.split(',')}
-     t_start = time.time()
-+    seed_r, micro = args.seed, 0
-     while step < args.steps:
-+        args.seed = seed_r + 1009*world*micro             # micro-step m: the streams of virtual rank r + m*world
-         try:
-             b = next(it)
-         except StopIteration:
-@@ -1255,8 +1267,33 @@ def main():
-                 l_adv = l_adv + args.adv_hidden_w*g_loss_masked(disc, fake, tokh)
-             w_adv = float(adaptive_weight(l_x0 + args.held_w*l_held, l_adv, rgb)) * args.adv_w
-             loss = loss + w_adv*l_adv
--        opt.zero_grad(set_to_none=True)
--        loss.backward()
-+        if micro == 0:
-+            opt.zero_grad(set_to_none=True)
-+        (loss/args.accum).backward()
-+        # Critic step after the last micro-step (the detached renders of this one do not depend on the generator step).
-+        # Its gradients are summed in d_buf (the generator backward also reaches the critic heads) and averaged over the
-+        # clips that took a critic step, over micro-steps and GPUs, so all ranks keep identical critic weights.
-+        if disc is not None and step >= args.adv_start:
-+            dps = list(disc.heads.parameters())
-+            if micro == 0:
-+                l_d, d_cnt, d_buf = 0., 0, [torch.zeros_like(p) for p in dps]
-+            if tokw is None or tokw.sum() > 0:
-+                ld = (d_loss(disc, target[0, judged], rgb[0, judged].detach().clamp(0, 1)) if tokw is None else
-+                      d_loss_masked(disc, target[0, judged], rgb[0, judged].detach().clamp(0, 1), tokw))
-+                if tokh is not None and tokh.sum() > 0:
-+                    ld = ld + args.adv_hidden_w*d_loss_masked(disc, target[0, judged], rgb[0, judged].detach().clamp(0, 1), tokh)
-+                if ot_fake is not None:                      # R1: off-trajectory renders are fakes too (reals: the window's frames)
-+                    ld = ld + d_loss(disc, target[0, :len(ot_fake)], ot_fake)
-+                for a_, g_ in zip(d_buf, torch.autograd.grad(ld, dps, allow_unused=True)):
-+                    if g_ is not None:
-+                        a_.add_(g_)
-+                d_cnt += 1; l_d = float(ld)
-+        else:
-+            l_d = 0.
-+        if micro < args.accum - 1:
-+            micro += 1
-+            continue
-+        args.seed, micro = seed_r, 0
-         if world > 1:                                       # average gradients over GPUs
-             params = [p for p in model.parameters() if p.requires_grad]
-             has = torch.tensor([p.grad is not None for p in params], device=dev, dtype=torch.uint8)
-@@ -1288,34 +1325,15 @@ def main():
-             grp['lr'] = lr*warm*(.1+.9*cos)
-         if not skipped:
-             opt.step()
--        l_d = 0.
--        d_ok = disc is not None and step >= args.adv_start and (tokw is None or tokw.sum() > 0)
--        if d_ok:
--            opt_d.zero_grad(set_to_none=True)
--            ld = (d_loss(disc, target[0, judged], rgb[0, judged].detach().clamp(0, 1)) if tokw is None else
--                  d_loss_masked(disc, target[0, judged], rgb[0, judged].detach().clamp(0, 1), tokw))
--            if tokh is not None and tokh.sum() > 0:
--                ld = ld + args.adv_hidden_w*d_loss_masked(disc, target[0, judged], rgb[0, judged].detach().clamp(0, 1), tokh)
--            if ot_fake is not None:                          # R1: off-trajectory renders are fakes too (reals: the window's frames)
--                ld = ld + d_loss(disc, target[0, :len(ot_fake)], ot_fake)
--            ld.backward(); l_d = float(ld)
--            if world == 1:
--                opt_d.step()
--        if world > 1 and disc is not None and step >= args.adv_start:
--            # Multi-GPU critic: every rank joins the all-reduce (ranks without a critic step contribute zero gradients) and
--            # the critic gradient is averaged over the ranks that took one, so all ranks keep identical critic weights.
--            if not d_ok:
--                opt_d.zero_grad(set_to_none=True)
--            dps = list(disc.heads.parameters())
--            for p in dps:
--                if p.grad is None:
--                    p.grad = torch.zeros_like(p)
--            n_ok = torch.tensor([float(d_ok)], device=dev); dist.all_reduce(n_ok)
--            flat = torch._utils._flatten_dense_tensors([p.grad for p in dps]); dist.all_reduce(flat)
-+        if disc is not None and step >= args.adv_start:
-+            n_ok = torch.tensor([float(d_cnt)], device=dev)
-+            flat = torch._utils._flatten_dense_tensors(d_buf)
-+            if world > 1:
-+                dist.all_reduce(n_ok); dist.all_reduce(flat)
-             if n_ok.item() > 0:
-                 flat /= n_ok.item()
--                for p, g_ in zip(dps, torch._utils._unflatten_dense_tensors(flat, [p.grad for p in dps])):
--                    p.grad.copy_(g_)
-+                for p, g_ in zip(dps, torch._utils._unflatten_dense_tensors(flat, d_buf)):
-+                    p.grad = g_.clone()
-                 opt_d.step()
-         with torch.no_grad():
-             d = min(.999, (1+step)/(10+step))

09_voxel_update_f1_harm/README.md

@@ -1,57 +0,0 @@
-# Voxel update head f1 (do no harm), step 6000: the head of the best orbit-benchmark arm
-
-`head_006000.pt` - the trained voxel update head (outputs/voxupd4/f1_harm/head_006000.pt) behind the **TSDF-decided write**,
-the best arm on the DL3DV orbit (revisit) benchmark as of 2026-10-30. The file holds the voxel head only (7.9 M parameters);
-the trunk, proposal head and surface decoder come from the clean voxel arm in `07_clean_voxel_f3r/`.
-
-- **Needs, besides this file** (run from the repo root):
-  - `07_clean_voxel_f3r/ema_070000.pt` at `outputs/clean_voxel_f3r/ema_070000.pt` (the base path this checkpoint records);
-  - the decoder construction files `free_completion_20260927/checkpoints/ray/step002000` from `03_checkpoints_streaming.zip`,
-    at `outputs/clean/free_completion_20260927/checkpoints/ray/step002000` (the path F3 records; the zip restores it under
-    `outputs/free_completion_20260927/`, so move or link it). F3's weights are loaded over it;
-  - the 3D-Belief prior `3d_belief_re10k_256.pt` at `$FLC_BELIEF_CKPT`, read when the trunk is built (F3's weights are loaded
-    over it). Not in this repo;
-  - the orbit caches `data/orbit_cache/dl3dv_val33` / `dl3dv_test33` (DL3DV orbits with VGGT-Omega geometry,
-    `scripts/data/build_orbit_cache.py`). Not in this repo.
-- **Training:** `flc/unified/train_voxel_update.py`, 6000 steps, one B200, about 1.6 h, from F3 70k with everything but the
-  voxel head frozen. Data: 400 DL3DV training orbits (`data/orbit_cache/dl3dv_train`), rollouts over each stream with the write
-  in the loop: A4 write (association along the ray 8 voxels, evidence weights, depth test 8, prune .01), evidence / age input.
-  Loss: L1 + LPIPS + 0.1 depth at window frames and at 5 other frames that see the window's region (up to 150 degrees away,
-  weight .7), plus the do-no-harm term (weight 3): a hinged penalty on any rise of the other frames' 16 px patch L1 / frame
-  LPIPS over the memory before the update. Full command: command.json.
-- **Code:** branch amodal-completion. Trained on 2026-10-29 from the working tree; the code is in commit 9edc5dd (2026-10-30),
-  whose later additions to the trainer are off by default (not re-run at that commit).
-- **Load** as an `eval_long_stream.py --snapshot` (dict with 'voxel_update', 'base', 'voxel_args', 'ema_head', 'args', 'step';
-  `flc.unified.train_root_ae.load_teacher`).
-
-## The best arm: TSDF-decided write
-
-Voxel memory written by this head. A wide-band TSDF (band 15 voxels and at least the merge range) is kept beside it, fed
-each window's own prediction and the observed colours. A window may change the memory only in its holes and in 16 px patches
-where the memory disagrees with the TSDF (log-depth > .4 on most covered pixels, or patch-mean colour > .3), about 21-24 % of
-its pixels; elsewhere the memory is kept as stored.
-
-    python scripts/unified/eval_long_stream.py --snapshot 09_voxel_update_f1_harm/head_006000.pt \
-      --data data/orbit_cache/dl3dv_test33 --clips 33 --loops 2 --async-period 0 --diag-drift --arm none --mem-rule voxel \
-      --vox-prune .01 --vox-assoc 8 --vox-wcap 1e9 --vox-evmax --vox-gate 8 \
-      --vox-patch 0 --vox-patch-ref tsdf --vox-patch-hole pixel --vox-patch-dd .4 --vox-patch-dc .3 \
-      --compose-tsdf 15 --compose-3d --out <dir>
-
-(from commit 9edc5dd, `--preset tsdf_decided` sets the write flags and `--compose-tsdf 15`.)
-
-dl3dv_test33 (33 DL3DV test orbits), the stream walked twice (6 orbits), the first orbit's held-out frames rendered after every
-window, all pixels (holes count). One seed, B200. Runs are not bit-reproducible (GPU nondeterminism flips borderline patch
-decisions, up to 0.09 dB per window on one scene).
-
-| test33, 6 orbits, all pixels | PSNR orbit 1 / 3 / 6 | change at 6 | LPIPS at 6 |
-|---|---|---|---|
-| headline A: memory + the TSDF's low frequencies (box-15 blur of TSDF - memory, TSDF in holes; `track_*_all_c15`) | 15.25 / 15.61 / 15.84 | +0.59 | .561 |
-| headline B: 3D recolouring (each root's Gaussians shifted by the mean TSDF - Gaussian colour; `track_*_all_c3r`) | 16.01 / 15.96 / 15.98 | -0.03 | .552 |
-| the memory alone (`track_*_all`) | 15.71 / 15.58 / 15.60 | -0.11 | .554 |
-| wide-band TSDF (baseline, same window predictions) | 14.98 / 15.36 / 15.56 | +0.58 | .622 |
-| this head, every write re-fuses (no TSDF decision) | 15.87 / 15.22 / 14.90 | -0.97 | .578 |
-
-Curves after every window: `best_arm_test33_6orbits.png` (numbers in `best_arm_test33_6orbits.json`). Which headline to report
-is still open.
-
-sha256 head_006000.pt: af728224d497462e686ae8275d6bc051a6adb7330e636808522a1ba100b38969

09_voxel_update_f1_harm/best_arm_test33_6orbits.json

@@ -1,72 +0,0 @@
-{
- "Headline A: memory + TSDF lows (c15)": {
-  "psnr": {
-   "n": 33,
-   "start": 11.810133934020996,
-   "mid": 15.608428868380459,
-   "end": 15.839989719968854
-  },
-  "lpips": {
-   "n": 33,
-   "start": 0.5832957029342651,
-   "mid": 0.563083928642851,
-   "end": 0.5604310893651211
-  }
- },
- "Headline B: 3D recolouring (c3r)": {
-  "psnr": {
-   "n": 33,
-   "start": 11.464308420817057,
-   "mid": 15.959378589283336,
-   "end": 15.981326305505002
-  },
-  "lpips": {
-   "n": 33,
-   "start": 0.5789229273796082,
-   "mid": 0.5521636569138729,
-   "end": 0.5524708339662263
-  }
- },
- "TSDF-decided memory alone": {
-  "psnr": {
-   "n": 33,
-   "start": 11.507584571838379,
-   "mid": 15.584838838288277,
-   "end": 15.585186756018436
-  },
-  "lpips": {
-   "n": 33,
-   "start": 0.5779265562693278,
-   "mid": 0.553575286359498,
-   "end": 0.5541323242765485
-  }
- },
- "Wide-band TSDF (baseline)": {
-  "psnr": {
-   "n": 33,
-   "start": 10.84955883026123,
-   "mid": 15.356085892879602,
-   "end": 15.56182676373106
-  },
-  "lpips": {
-   "n": 33,
-   "start": 0.6679924329121908,
-   "mid": 0.6317940372409243,
-   "end": 0.6224375312978571
-  }
- },
- "f1 head, every write re-fuses": {
-  "psnr": {
-   "n": 33,
-   "start": 11.507584571838379,
-   "mid": 15.215329430320047,
-   "end": 14.902457555135092
-  },
-  "lpips": {
-   "n": 33,
-   "start": 0.5779262781143188,
-   "mid": 0.5684382572318568,
-   "end": 0.5781347408439174
-  }
- }
-}

09_voxel_update_f1_harm/command.json

@@ -1,75 +0,0 @@
-{
- "argv": [
-  "flc/unified/train_voxel_update.py",
-  "--out",
-  "outputs/voxupd4/f1_harm",
-  "--snapshot",
-  "outputs/clean_voxel_f3r/ema_070000.pt",
-  "--data",
-  "data/orbit_cache/dl3dv_train",
-  "--steps",
-  "6000",
-  "--save-every",
-  "1500",
-  "--evidence",
-  "--prune",
-  ".01",
-  "--max-ang",
-  "150",
-  "--see-region",
-  "--assoc",
-  "8",
-  "--evw",
-  "--gate",
-  "8",
-  "--w-harm",
-  "3"
- ],
- "args": {
-  "out": "outputs/voxupd4/f1_harm",
-  "snapshot": "outputs/clean_voxel_f3r/ema_070000.pt",
-  "data": "data/orbit_cache/dl3dv_train",
-  "overfit": "",
-  "exclude": "",
-  "steps": "6000",
-  "rollouts": "2",
-  "lr": "5e-05",
-  "warmup": "100",
-  "ema": "0.999",
-  "n_win": "3",
-  "n_other": "5",
-  "far": "12",
-  "max_ang": "150.0",
-  "see_region": "True",
-  "w_other": "0.7",
-  "lpips_w": "1.0",
-  "w_harm": "3.0",
-  "harm_patch": "16",
-  "w_depth": "0.1",
-  "dep_clamp": "0.3",
-  "hold": "False",
-  "wcap": "0.0",
-  "evidence": "True",
-  "gate": "8.0",
-  "band": "False",
-  "evw": "True",
-  "assoc": "8.0",
-  "assoc_lat": "1.0",
-  "accum": "",
-  "accum_range": "8.0",
-  "angle": "False",
-  "train_trunk": "False",
-  "lr_trunk": "2e-05",
-  "w_self": "0.2",
-  "prune": "0.01",
-  "max_roots_render": "200000",
-  "log_every": "20",
-  "save_every": "1500",
-  "ot_w": "0.0",
-  "ot_geo_w": "0.0",
-  "ot_n": "2",
-  "ot_frac": "0.3",
-  "adv_start": "300",
-  "seed": "0"
- }
-}

09_voxel_update_f1_harm/head_006000.pt

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:af728224d497462e686ae8275d6bc051a6adb7330e636808522a1ba100b38969
-size 31681221

10_fusion_head_frozen_3k/README.md

@@ -1,50 +0,0 @@
-# Fusion head, decoder frozen, EMA at step 3000: learned decoding of the feature-fusion memory
-
-`fhead_003000.pt` - the fusion head (outputs/fusion_head/fh_frozen/fhead_003000.pt) of the feature-fusion memory: a per-root
-correction of the fused state before the frozen surface decoder renders it. 1.86 M parameters; the file holds the EMA head
-only (`decoder` is None: the surface decoder is F3's, unchanged).
-
-- **The memory it decodes** (`flc/unified/fused_memory.py`, `eval_long_stream.py --memory fused`): the state is averaged
-  evidence, never a decoded output. A wide-band TSDF (band 15 voxels, at least .8) lifts each window's proposal fields per pixel
-  (feature 256, frame / content vectors, extent, activity) as running means; one root per 0.1 cell of TSDF surface (position,
-  normal and the decoder's chart from the TSDF, the other fields averaged); an 8 x 8 x 16 appearance grid per root averaged from
-  each new frame's appearance map. The window model is the f1 snapshot (`09_voxel_update_f1_harm/`: trunk and proposal head).
-- **The head** (`flc/unified/fusion_head.py`): input = the fused fields in the root's frame, the grid and 7 TSDF statistics;
-  attention over the 26 neighbouring cells; zero-initialised corrections (the identity at step 0) of feature, content vectors,
-  extent, activity, every grid cell and the position along the normal (at most half a cell).
-- **Needs, besides this file** (run from the repo root): `09_voxel_update_f1_harm/head_006000.pt` (passed as `--snapshot`) and
-  everything that file needs (F3 `07_clean_voxel_f3r/`, the decoder construction files from `03_checkpoints_streaming.zip`, the
-  3D-Belief prior, the orbit caches; see that README).
-- **Training:** `flc/unified/train_fusion_head.py` (command.json), 3000 of 4000 steps, one B200, ~3.7 s/step, decoder frozen.
-  Rollouts over 400 DL3DV training orbits (`data/orbit_cache/dl3dv_train`), each walked twice; the fused state does not depend
-  on the head, so there is no on-policy drift. Targets per step: 3 window frames, 4 observed frames of other passes, 2 held-out
-  frames (within 150 degrees, seeing the window's region) and 2 held-out frames inside the window's span. Loss: L1 + LPIPS on
-  all pixels + 0.1 x clamped log-depth L1. AdamW lr 1e-4, warmup 200 + cosine to 4000, EMA .999, seed 0.
-- **Code:** branch amodal-completion, commit 9edc5dd (the code trained and evaluated here, unchanged since launch).
-- **Load:** `FusionHead(**ck['head_args'])` + `load_state_dict(ck['fusion_head'])` (`flc/unified/fusion_head.py`), as
-  `eval_long_stream.py --fusion-head` does.
-
-## Evaluation
-
-    python scripts/unified/eval_long_stream.py --snapshot 09_voxel_update_f1_harm/head_006000.pt --memory fused \
-      --fusion-head 10_fusion_head_frozen_3k/fhead_003000.pt --data data/orbit_cache/dl3dv_val33 --clips 33 --loops 2 \
-      --async-period 0 --out <dir>
-
-dl3dv_val33 (33 DL3DV orbits unused in training), the stream walked twice (6 orbits), the first orbit's held-out frames
-rendered after every window, all pixels (holes count). One seed, B200. **Validation slice only: not yet run on dl3dv_test33.**
-
-| val33, 6 orbits, all pixels | PSNR orbit 1 / 3 / 6 | change at 6 | LPIPS at 6 | stored per scene |
-|---|---|---|---|---|
-| **fused memory + this head** | 14.84 / 14.65 / 15.07 | +0.24 | .562 | ~11.5 GB |
-| same, decoder unlocked (lr 1e-5; not uploaded) | 14.81 / 14.63 / 15.08 | +0.26 | .562 | ~11.5 GB |
-| fused memory decoded as is (no head) | 13.95 / 13.84 / 14.71 | +0.76 | .637 | ~11.5 GB |
-| best arm (`09_voxel_update_f1_harm/`), headline A: composite | 14.65 / 14.80 / 15.23 | +0.57 | .565 | ~0.5 GB |
-| best arm, headline B: 3D recolouring | 15.37 / 15.21 / 15.23 | -0.14 | .562 | ~0.5 GB |
-| wide-band TSDF | 14.17 / 14.34 / 14.69 | +0.52 | .621 | 0.24 GB |
-
-The head reaches the best arm's sharpness as one 3D state (no image-space blend) and still rises over the stream, but ends
-0.16 dB below both best-arm headlines and is 0.5 dB below headline B at orbit 1. Curves after every window:
-`fusion_head_val33_6orbits.png` (numbers in `fusion_head_val33_6orbits.json`). The ~11.5 GB per scene (lifted features kept on
-every wide-band TSDF voxel, fp32 appearance sums for every cell ever seen) is a storage choice of this version, not of the head.
-
-sha256 fhead_003000.pt: b8e2a97a7557a30c12cca2fee04d1fa3d9ad713f74b7684d73b909b1e6b34614

10_fusion_head_frozen_3k/command.json

@@ -1,39 +0,0 @@
-{
- "argv": [
-  "flc/unified/train_fusion_head.py",
-  "--out",
-  "outputs/fusion_head/fh_frozen",
-  "--steps",
-  "4000",
-  "--save-every",
-  "500"
- ],
- "args": {
-  "out": "outputs/fusion_head/fh_frozen",
-  "snapshot": "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/free-latent-completion/outputs/voxupd4/f1_harm/head_006000.pt",
-  "data": "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/free-latent-completion/data/orbit_cache/dl3dv_train",
-  "steps": "4000",
-  "rollouts": "2",
-  "loops": "2",
-  "lr": "0.0001",
-  "dec_lr": "0.0",
-  "warmup": "200",
-  "ema": "0.999",
-  "n_win": "3",
-  "n_other": "4",
-  "n_held": "2",
-  "n_held_near": "2",
-  "far": "12",
-  "max_ang": "150.0",
-  "lpips_w": "1.0",
-  "w_depth": "0.1",
-  "dep_clamp": "0.3",
-  "log_every": "20",
-  "save_every": "500",
-  "seed": "0",
-  "see_region": "True",
-  "w_tsdf": "0.0",
-  "w_tsdf_dep": "0.0",
-  "tsdf_n": "0"
- }
-}

10_fusion_head_frozen_3k/fhead_003000.pt

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:b8e2a97a7557a30c12cca2fee04d1fa3d9ad713f74b7684d73b909b1e6b34614
-size 7471341

10_fusion_head_frozen_3k/fusion_head_val33_6orbits.json

@@ -1,86 +0,0 @@
-{
- "Best arm, headline A (c15)": {
-  "psnr": {
-   "n": 33,
-   "start": 11.419811248779297,
-   "mid": 14.797905488447709,
-   "end": 15.225642435478441
-  },
-  "lpips": {
-   "n": 33,
-   "start": 0.6384534537792206,
-   "mid": 0.5725538811900399,
-   "end": 0.5654349832823782
-  }
- },
- "Best arm, headline B (c3r)": {
-  "psnr": {
-   "n": 33,
-   "start": 11.319180011749268,
-   "mid": 15.206268194949988,
-   "end": 15.226465080723617
-  },
-  "lpips": {
-   "n": 33,
-   "start": 0.6169425845146179,
-   "mid": 0.5620349013444149,
-   "end": 0.5619459242531748
-  }
- },
- "Wide-band TSDF (baseline)": {
-  "psnr": {
-   "n": 33,
-   "start": 9.873345375061035,
-   "mid": 14.339310357064912,
-   "end": 14.690505056670219
-  },
-  "lpips": {
-   "n": 33,
-   "start": 0.6851010620594025,
-   "mid": 0.6363881736090689,
-   "end": 0.6210566632675402
-  }
- },
- "Fused memory, decoded as is": {
-  "psnr": {
-   "n": 33,
-   "start": 10.664002418518066,
-   "mid": 13.844639749237984,
-   "end": 14.713996945005475
-  },
-  "lpips": {
-   "n": 33,
-   "start": 0.7107486128807068,
-   "mid": 0.6630020141601562,
-   "end": 0.6371301557078506
-  }
- },
- "Fused memory + fusion head 3k, decoder frozen": {
-  "psnr": {
-   "n": 33,
-   "start": 10.80231237411499,
-   "mid": 14.645491975726504,
-   "end": 15.073678507949367
-  },
-  "lpips": {
-   "n": 33,
-   "start": 0.6445749998092651,
-   "mid": 0.5773161803231095,
-   "end": 0.5619311404950691
-  }
- },
- "Fused memory + fusion head 3k, decoder unlocked": {
-  "psnr": {
-   "n": 33,
-   "start": 10.772018909454346,
-   "mid": 14.634032625140566,
-   "end": 15.076873923792984
-  },
-  "lpips": {
-   "n": 33,
-   "start": 0.6380249559879303,
-   "mid": 0.576788578972672,
-   "end": 0.5615499796289386
-  }
- }
-}

14_stream_fixer_s200_paired/README.md

@@ -1,25 +0,0 @@
-# 14_stream_fixer_s200_paired
-
-The frozen render fixer that sits in the loop of every Gen3R round-2 experiment and of our streaming memory
-("s200_paired"): a one-step diffusion fixer (SyncFix) fine-tuned on our own memory's renders.
-
-| file | restore to | size |
-|---|---|---|
-| `denoiser_004000.pt` | `outputs/syncfix_ft/s200_paired/denoiser_004000.pt` | 4.6 GB |
-| `args.json`, `log.jsonl`, `train.log` | same folder | small |
-
-**Lineage.** SyncFix official weights (github.com/Deming77/SyncFix; the checkpoint is a Google Drive download listed in that
-README) → `outputs/syncfix_ft/a2_res_w10/denoiser_008000.pt` (8,000 steps on our renders) → this file (4,000 more steps, lr 1e-5,
-batch 2, paired + cross-view agreement losses on the stream bank `outputs/long_stream/best_fixbank_train3/bank`; `args.json`). The
-file holds the full fine-tuned network, so only the SyncFix code is needed to run it, not the official checkpoint.
-
-**Use.**
-- Online in the memory: `scripts/unified/eval_long_stream.py ... --ext-fix syncfix:outputs/syncfix_ft/s200_paired/denoiser_004000.pt`.
-- Gen3R round-2 probe: `scripts/lyra/teacher_probe2.py` (fix stage) makes Gen3R's conditioning frame and SDEdit start from it.
-- Initialisation of the output fixers in `15_output_fixers/`.
-
-**Results** (docs/RESULTS.md §0r): on the val33 stream bank (8,546 views) 17.47 → 17.56 dB, LPIPS .324 → .313, sharpness 0.21 →
-0.66 of real. In the Gen3R round-2 probe the fixed render beat every Gen3R variant (RESULTS §0q).
-
-**Terms.** Derived from SyncFix, whose repository has no license file; released here on the project owner's decision (2026-11-04).
-Check SyncFix's terms before reuse. Trained on renders of DL3DV scenes (DL3DV-10K terms of use).

14_stream_fixer_s200_paired/SHA256SUMS

@@ -1,5 +0,0 @@
-07cacdfed9423580d7dd38174d263f5b5e8949fa6748f2d6d72cd9a4e9e14485  ./args.json
-cf7144c53166292f57ebad33915a5310b1aecc3dff471e82938866a9c1469528  ./denoiser_004000.pt
-3eb06516986d2e7db8331672c0aaebb061a407334afb415861ef252627b16c4e  ./log.jsonl
-da6f8d29784c6a0f6aee80504c8e9326cf36d0f8fb7bc80c1b8498c4a411db9d  ./README.md
-190b13a650307d4b6153a9698bf83f12fce0baf63222930b61b8721cb8d27d6c  ./train.log

14_stream_fixer_s200_paired/args.json

@@ -1,21 +0,0 @@
-{
- "banks": "[PosixPath('outputs/long_stream/best_fixbank_train3/bank')]",
- "n_orbits": "0",
- "val_bank": "outputs/long_stream/best_val33/bank_near",
- "out": "outputs/syncfix_ft/s200_paired",
- "init": "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/free-latent-completion/outputs/syncfix_ft/a2_res_w10/denoiser_008000.pt",
- "w_agree": "10.0",
- "w_near": "0.0",
- "w_app": "0.5",
- "w_geo": "0.5",
- "realism_warmup": "500",
- "steps": "4000",
- "lr": "1e-05",
- "lr_warmup": "200",
- "batch": "2",
- "val_every": "500",
- "val_windows": "48",
- "save_every": "2000",
- "workers": "6",
- "seed": "1102"
-}

14_stream_fixer_s200_paired/denoiser_004000.pt

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:cf7144c53166292f57ebad33915a5310b1aecc3dff471e82938866a9c1469528
-size 4853065399

14_stream_fixer_s200_paired/log.jsonl

@@ -1,89 +0,0 @@
-{"val": {"psnr_r": 17.316714773575466, "psnr_f": 16.838129301865894, "lpips_r": 0.3291937040630728, "lpips_f": 0.3579407640111943, "sharp_r": 0.20774495438672602, "sharp_f": 0.6178478267975152, "agree": 0.195467301333944, "struct": null, "app": null, "geo": null, "step": 0}}
-{"step": 50, "loss": 5.282959461212158, "gnorm": 19.012298583984375, "s_per_step": 0.46517508029937743, "lat": 0.343128502368927, "pix": 3.6950299739837646, "agree": 0.12448009848594666}
-{"step": 100, "loss": 5.966648101806641, "gnorm": 19.931276321411133, "s_per_step": 0.45174971103668216, "lat": 0.6359272003173828, "pix": 3.6033713817596436, "agree": 0.17273494601249695}
-{"step": 150, "loss": 6.706423282623291, "gnorm": 20.032516479492188, "s_per_step": 0.44838563601175946, "lat": 0.4129915237426758, "pix": 4.5263471603393555, "agree": 0.17670845985412598}
-{"step": 200, "loss": 5.409975051879883, "gnorm": 17.286827087402344, "s_per_step": 0.444759224653244, "lat": 0.5236479640007019, "pix": 3.4860780239105225, "agree": 0.14002491533756256}
-{"step": 250, "loss": 2.517254114151001, "gnorm": 9.617813110351562, "s_per_step": 0.44287422657012937, "lat": 0.12476496398448944, "pix": 1.932403564453125, "agree": 0.046008553355932236}
-{"step": 300, "loss": 4.430167198181152, "gnorm": 11.956856727600098, "s_per_step": 0.44164979934692383, "lat": 0.3534668982028961, "pix": 2.9653356075286865, "agree": 0.11113643646240234}
-{"step": 350, "loss": 4.688089370727539, "gnorm": 15.709254264831543, "s_per_step": 0.44050958156585696, "lat": 0.3752760589122772, "pix": 3.085775375366211, "agree": 0.12270380556583405}
-{"step": 400, "loss": 7.314797401428223, "gnorm": 20.28812026977539, "s_per_step": 0.43988965809345243, "lat": 0.6609450578689575, "pix": 4.753818511962891, "agree": 0.1900033950805664}
-{"step": 450, "loss": 2.8827595710754395, "gnorm": 7.827585220336914, "s_per_step": 0.4389246956507365, "lat": 0.12322597205638885, "pix": 2.1229665279388428, "agree": 0.06365671753883362}
-{"step": 500, "loss": 5.692932605743408, "gnorm": 11.124910354614258, "s_per_step": 0.4382571859359741, "lat": 0.5842304229736328, "pix": 3.371189832687378, "agree": 0.17375124990940094}
-{"val": {"psnr_r": 17.316714773575466, "psnr_f": 17.256229897340138, "lpips_r": 0.3291937040630728, "lpips_f": 0.3277533127305408, "sharp_r": 0.20774495438672602, "sharp_f": 0.6295974360158046, "agree": 0.19425605920453867, "struct": null, "app": null, "geo": null, "step": 500}}
-{"step": 550, "loss": 5.257030963897705, "gnorm": 15.131274223327637, "s_per_step": 0.44431875185533004, "lat": 0.4374696612358093, "pix": 3.6632354259490967, "agree": 0.11563259363174438}
-{"step": 600, "loss": 7.337639808654785, "gnorm": 19.91976547241211, "s_per_step": 0.44358083645502727, "lat": 0.6734758615493774, "pix": 5.058652877807617, "agree": 0.16055111587047577}
-{"step": 650, "loss": 5.58404016494751, "gnorm": 36.128883361816406, "s_per_step": 0.4425516605377197, "lat": 0.406619131565094, "pix": 3.679435968399048, "agree": 0.14979849755764008}
-{"step": 700, "loss": 6.593929290771484, "gnorm": 20.724973678588867, "s_per_step": 0.44170593397957936, "lat": 0.5818449258804321, "pix": 4.306272029876709, "agree": 0.17058126628398895}
-{"step": 750, "loss": 4.463681221008301, "gnorm": 14.081138610839844, "s_per_step": 0.4409639717737834, "lat": 0.3447449505329132, "pix": 3.111719846725464, "agree": 0.10072162747383118}
-{"step": 800, "loss": 4.976413726806641, "gnorm": 18.86370277404785, "s_per_step": 0.44031798243522646, "lat": 0.7762983441352844, "pix": 3.1152749061584473, "agree": 0.10848403722047806}
-{"step": 850, "loss": 3.5793182849884033, "gnorm": 11.016100883483887, "s_per_step": 0.4396640449411729, "lat": 0.631940484046936, "pix": 2.4261178970336914, "agree": 0.05212600529193878}
-{"step": 900, "loss": 4.1083269119262695, "gnorm": 17.977937698364258, "s_per_step": 0.4393557726012336, "lat": 0.32825568318367004, "pix": 2.822862386703491, "agree": 0.09572090208530426}
-{"step": 950, "loss": 3.434788703918457, "gnorm": 6.248925685882568, "s_per_step": 0.4390755520368877, "lat": 0.3318836987018585, "pix": 2.2599053382873535, "agree": 0.0842999666929245}
-{"step": 1000, "loss": 3.451570987701416, "gnorm": 9.025375366210938, "s_per_step": 0.4386509702205658, "lat": 0.31044721603393555, "pix": 2.3478565216064453, "agree": 0.07932672649621964}
-{"val": {"psnr_r": 17.316714773575466, "psnr_f": 17.283803939819336, "lpips_r": 0.3291937040630728, "lpips_f": 0.3256644190599521, "sharp_r": 0.20774495438672602, "sharp_f": 0.6612661257386208, "agree": 0.19192229863256216, "struct": null, "app": null, "geo": null, "step": 1000}}
-{"step": 1050, "loss": 3.5715818405151367, "gnorm": 14.426520347595215, "s_per_step": 0.4416197402136666, "lat": 0.3895701467990875, "pix": 2.0408735275268555, "agree": 0.11411382257938385}
-{"step": 1100, "loss": 4.559601306915283, "gnorm": 8.70854663848877, "s_per_step": 0.4411098217964172, "lat": 0.47088319063186646, "pix": 3.1535019874572754, "agree": 0.09352162480354309}
-{"step": 1150, "loss": 5.932168006896973, "gnorm": 18.10098648071289, "s_per_step": 0.4406720725349758, "lat": 0.4766201674938202, "pix": 3.861445188522339, "agree": 0.1594102680683136}
-{"step": 1200, "loss": 2.7961795330047607, "gnorm": 9.36918830871582, "s_per_step": 0.44051453212896985, "lat": 0.484896183013916, "pix": 1.8497220277786255, "agree": 0.0461561493575573}
-{"step": 1250, "loss": 3.6632070541381836, "gnorm": 39.42759704589844, "s_per_step": 0.44023946323394775, "lat": 0.315992534160614, "pix": 2.6957170963287354, "agree": 0.06514973938465118}
-{"step": 1300, "loss": 2.7233002185821533, "gnorm": 8.128558158874512, "s_per_step": 0.43990861415863036, "lat": 0.11197896301746368, "pix": 1.9066317081451416, "agree": 0.0704689547419548}
-{"step": 1350, "loss": 6.2007317543029785, "gnorm": 7.34594202041626, "s_per_step": 0.439667704017074, "lat": 0.5812610387802124, "pix": 4.5055365562438965, "agree": 0.11139339208602905}
-{"step": 1400, "loss": 1.6749701499938965, "gnorm": 29.470783233642578, "s_per_step": 0.43953941924231393, "lat": 0.08110769093036652, "pix": 1.2208553552627563, "agree": 0.037300702184438705}
-{"step": 1450, "loss": 5.254883289337158, "gnorm": 17.242197036743164, "s_per_step": 0.4395118335197712, "lat": 0.6751502752304077, "pix": 3.398437738418579, "agree": 0.11812955141067505}
-{"step": 1500, "loss": 2.305231809616089, "gnorm": 7.868963718414307, "s_per_step": 0.43940510670344035, "lat": 0.10623176395893097, "pix": 1.6000741720199585, "agree": 0.05989258363842964}
-{"val": {"psnr_r": 17.316714773575466, "psnr_f": 17.341512391964596, "lpips_r": 0.3291937040630728, "lpips_f": 0.3219629074446857, "sharp_r": 0.20774495438672602, "sharp_f": 0.6501046052823464, "agree": 0.1908291537935535, "struct": null, "app": null, "geo": null, "step": 1500}}
-{"step": 1550, "loss": 4.3534464836120605, "gnorm": 12.095688819885254, "s_per_step": 0.44154483749020484, "lat": 0.4174262285232544, "pix": 2.748300075531006, "agree": 0.11877201497554779}
-{"step": 1600, "loss": 4.407609462738037, "gnorm": 15.485705375671387, "s_per_step": 0.4413210020959377, "lat": 0.417605996131897, "pix": 3.133139133453369, "agree": 0.08568646013736725}
-{"step": 1650, "loss": 4.763701915740967, "gnorm": 15.481014251708984, "s_per_step": 0.441193803729433, "lat": 0.7621040940284729, "pix": 3.054715633392334, "agree": 0.09468822181224823}
-{"step": 1700, "loss": 3.5829763412475586, "gnorm": 9.605074882507324, "s_per_step": 0.4410405219302458, "lat": 0.40083324909210205, "pix": 2.3861889839172363, "agree": 0.0795954167842865}
-{"step": 1750, "loss": 2.658104419708252, "gnorm": 26.264644622802734, "s_per_step": 0.44086836310795374, "lat": 0.3701430559158325, "pix": 1.8756552934646606, "agree": 0.04123059660196304}
-{"step": 1800, "loss": 4.382001876831055, "gnorm": 10.638790130615234, "s_per_step": 0.4407528383202023, "lat": 0.15204757452011108, "pix": 2.7892963886260986, "agree": 0.1440657675266266}
-{"step": 1850, "loss": 5.236424446105957, "gnorm": 9.37230396270752, "s_per_step": 0.44063499450683596, "lat": 0.33061152696609497, "pix": 3.796599864959717, "agree": 0.11092129349708557}
-{"step": 1900, "loss": 3.8434321880340576, "gnorm": 12.531085968017578, "s_per_step": 0.4404499878381428, "lat": 0.30072134733200073, "pix": 2.5734760761260986, "agree": 0.09692347794771194}
-{"step": 1950, "loss": 4.5669779777526855, "gnorm": 12.073986053466797, "s_per_step": 0.4402804494515443, "lat": 0.46053624153137207, "pix": 3.212953567504883, "agree": 0.08934880048036575}
-{"step": 2000, "loss": 2.833603858947754, "gnorm": 7.930850028991699, "s_per_step": 0.4401915755271912, "lat": 0.32637524604797363, "pix": 1.9994274377822876, "agree": 0.050780102610588074}
-{"val": {"psnr_r": 17.316714773575466, "psnr_f": 17.34564033150673, "lpips_r": 0.3291937040630728, "lpips_f": 0.31796501437202096, "sharp_r": 0.20774495438672602, "sharp_f": 0.645380032248795, "agree": 0.19199531494329372, "struct": null, "app": null, "geo": null, "step": 2000}}
-{"step": 2050, "loss": 5.759422779083252, "gnorm": 14.33885669708252, "s_per_step": 0.4463263610514199, "lat": 0.4326610565185547, "pix": 3.841921329498291, "agree": 0.1484840363264084}
-{"step": 2100, "loss": 3.200852870941162, "gnorm": 6.449644088745117, "s_per_step": 0.4461822601727077, "lat": 0.2919710874557495, "pix": 2.1164474487304688, "agree": 0.07924340665340424}
-{"step": 2150, "loss": 2.6471176147460938, "gnorm": 8.536527633666992, "s_per_step": 0.44592188624448553, "lat": 0.08426481485366821, "pix": 1.961434006690979, "agree": 0.06014186143875122}
-{"step": 2200, "loss": 2.8024680614471436, "gnorm": 5.264009952545166, "s_per_step": 0.4456899069656025, "lat": 0.2832692563533783, "pix": 1.9454680681228638, "agree": 0.05737307295203209}
-{"step": 2250, "loss": 3.7711868286132812, "gnorm": 10.562610626220703, "s_per_step": 0.44545665062798395, "lat": 0.1016571968793869, "pix": 2.797351360321045, "agree": 0.08721782267093658}
-{"step": 2300, "loss": 5.900383949279785, "gnorm": 32.107784271240234, "s_per_step": 0.4451768103889797, "lat": 0.4457588791847229, "pix": 3.7615573406219482, "agree": 0.16930678486824036}
-{"step": 2350, "loss": 1.7644014358520508, "gnorm": 8.211969375610352, "s_per_step": 0.44487129870881426, "lat": 0.21069106459617615, "pix": 1.1989392042160034, "agree": 0.03547711670398712}
-{"step": 2400, "loss": 2.5445470809936523, "gnorm": 11.1336669921875, "s_per_step": 0.4446574640274048, "lat": 0.11066100001335144, "pix": 1.7327866554260254, "agree": 0.07010994851589203}
-{"step": 2450, "loss": 4.296110153198242, "gnorm": 17.02275276184082, "s_per_step": 0.444463370186942, "lat": 0.615707516670227, "pix": 2.907406806945801, "agree": 0.07729961723089218}
-{"step": 2500, "loss": 3.719048500061035, "gnorm": 29.368309020996094, "s_per_step": 0.4442202063560486, "lat": 0.3668941855430603, "pix": 2.6853716373443604, "agree": 0.0666782557964325}
-{"val": {"psnr_r": 17.316714773575466, "psnr_f": 17.416791707277298, "lpips_r": 0.3291937040630728, "lpips_f": 0.3185555483214557, "sharp_r": 0.20774495438672602, "sharp_f": 0.6460029010971388, "agree": 0.1897952084740003, "struct": null, "app": null, "geo": null, "step": 2500}}
-{"step": 2550, "loss": 3.2296857833862305, "gnorm": 13.614500045776367, "s_per_step": 0.44535896441515754, "lat": 0.5099461078643799, "pix": 2.08665132522583, "agree": 0.06330884248018265}
-{"step": 2600, "loss": 2.702807903289795, "gnorm": 8.828268051147461, "s_per_step": 0.44510721637652473, "lat": 0.30793821811676025, "pix": 1.9307066202163696, "agree": 0.04641629382967949}
-{"step": 2650, "loss": 2.958904504776001, "gnorm": 7.468323707580566, "s_per_step": 0.44493436552443594, "lat": 0.34191662073135376, "pix": 1.9293924570083618, "agree": 0.06875953823328018}
-{"step": 2700, "loss": 4.413849353790283, "gnorm": 19.28907585144043, "s_per_step": 0.4447533882988824, "lat": 0.3250659108161926, "pix": 3.1127989292144775, "agree": 0.09759844839572906}
-{"step": 2750, "loss": 5.2409257888793945, "gnorm": 33.169105529785156, "s_per_step": 0.44459822247245095, "lat": 0.39315852522850037, "pix": 3.3126258850097656, "agree": 0.15351416170597076}
-{"step": 2800, "loss": 4.302533149719238, "gnorm": 11.56947135925293, "s_per_step": 0.4444024190732411, "lat": 0.15460963547229767, "pix": 2.7358012199401855, "agree": 0.14121223986148834}
-{"step": 2850, "loss": 3.6289782524108887, "gnorm": 11.843771934509277, "s_per_step": 0.44424590955700793, "lat": 0.3586495518684387, "pix": 2.2874715328216553, "agree": 0.0982857197523117}
-{"step": 2900, "loss": 1.6281397342681885, "gnorm": 15.552648544311523, "s_per_step": 0.44409301058999423, "lat": 0.07965855300426483, "pix": 1.0718249082565308, "agree": 0.047665636986494064}
-{"step": 2950, "loss": 3.4574906826019287, "gnorm": 7.93548583984375, "s_per_step": 0.4439588007684481, "lat": 0.1403726041316986, "pix": 2.7158212661743164, "agree": 0.06012969836592674}
-{"step": 3000, "loss": 3.342578887939453, "gnorm": 13.835026741027832, "s_per_step": 0.443789825518926, "lat": 0.5397420525550842, "pix": 2.0133090019226074, "agree": 0.07895278930664062}
-{"val": {"psnr_r": 17.316714773575466, "psnr_f": 17.34444770216942, "lpips_r": 0.3291937040630728, "lpips_f": 0.3188491778758665, "sharp_r": 0.20774495438672602, "sharp_f": 0.7042727932954828, "agree": 0.190136032178998, "struct": null, "app": null, "geo": null, "step": 3000}}
-{"step": 3050, "loss": 5.019059181213379, "gnorm": 15.61850643157959, "s_per_step": 0.44479312912362523, "lat": 0.5351306200027466, "pix": 3.1667582988739014, "agree": 0.1317170262336731}
-{"step": 3100, "loss": 4.591529369354248, "gnorm": 12.626097679138184, "s_per_step": 0.4446433416489632, "lat": 0.3467460870742798, "pix": 3.2406888008117676, "agree": 0.10040946304798126}
-{"step": 3150, "loss": 2.716024160385132, "gnorm": 11.156320571899414, "s_per_step": 0.4444743940565321, "lat": 0.34542331099510193, "pix": 1.8316679000854492, "agree": 0.05389329791069031}
-{"step": 3200, "loss": 2.562666893005371, "gnorm": 14.974357604980469, "s_per_step": 0.4442821995168924, "lat": 0.3057880997657776, "pix": 1.7902412414550781, "agree": 0.04666374996304512}
-{"step": 3250, "loss": 3.2485222816467285, "gnorm": 56.41815948486328, "s_per_step": 0.44420252623924844, "lat": 0.32220369577407837, "pix": 2.32841420173645, "agree": 0.05979045107960701}
-{"step": 3300, "loss": 2.6899020671844482, "gnorm": 8.662028312683105, "s_per_step": 0.44408551981954864, "lat": 0.32022303342819214, "pix": 1.8101767301559448, "agree": 0.05595023185014725}
-{"step": 3350, "loss": 2.355391502380371, "gnorm": 8.354748725891113, "s_per_step": 0.4439414439272525, "lat": 0.08787134289741516, "pix": 1.7305134534835815, "agree": 0.05370066314935684}
-{"step": 3400, "loss": 2.1708486080169678, "gnorm": 7.437104225158691, "s_per_step": 0.4437864069377675, "lat": 0.12262387573719025, "pix": 1.7171765565872192, "agree": 0.03310480713844299}
-{"step": 3450, "loss": 5.309618949890137, "gnorm": 12.234618186950684, "s_per_step": 0.44364174144855445, "lat": 0.5908336639404297, "pix": 3.537775993347168, "agree": 0.11810091882944107}
-{"step": 3500, "loss": 4.368440628051758, "gnorm": 11.99641227722168, "s_per_step": 0.4434744747025626, "lat": 0.42390772700309753, "pix": 2.8771071434020996, "agree": 0.10674256086349487}
-{"val": {"psnr_r": 17.316714773575466, "psnr_f": 17.351207494735718, "lpips_r": 0.3291937040630728, "lpips_f": 0.3177070333622396, "sharp_r": 0.20774495438672602, "sharp_f": 0.7195128429060181, "agree": 0.19155894871801138, "struct": null, "app": null, "geo": null, "step": 3500}}
-{"step": 3550, "loss": 3.2589502334594727, "gnorm": 7.004642963409424, "s_per_step": 0.44433776170435085, "lat": 0.33430784940719604, "pix": 2.292813301086426, "agree": 0.06318289786577225}
-{"step": 3600, "loss": 2.950033187866211, "gnorm": 5.723316192626953, "s_per_step": 0.44417368829250337, "lat": 0.14182040095329285, "pix": 2.017871379852295, "agree": 0.07903414964675903}
-{"step": 3650, "loss": 2.969468116760254, "gnorm": 6.427953720092773, "s_per_step": 0.4439964721627431, "lat": 0.3132466673851013, "pix": 1.9700448513031006, "agree": 0.06861765682697296}
-{"step": 3700, "loss": 2.945732831954956, "gnorm": 15.645040512084961, "s_per_step": 0.44382294925483495, "lat": 0.12360095232725143, "pix": 2.2259063720703125, "agree": 0.05962254852056503}
-{"step": 3750, "loss": 4.750136852264404, "gnorm": 10.02662181854248, "s_per_step": 0.44365882555643715, "lat": 0.37855973839759827, "pix": 3.265590190887451, "agree": 0.11059869080781937}
-{"step": 3800, "loss": 5.50206184387207, "gnorm": 15.832613945007324, "s_per_step": 0.4435086402767583, "lat": 0.48081135749816895, "pix": 3.5999038219451904, "agree": 0.1421346366405487}
-{"step": 3850, "loss": 4.275483131408691, "gnorm": 11.54232406616211, "s_per_step": 0.443364114451718, "lat": 0.30817466974258423, "pix": 3.0791637897491455, "agree": 0.08881445229053497}
-{"step": 3900, "loss": 2.1928091049194336, "gnorm": 8.542862892150879, "s_per_step": 0.4432803163772974, "lat": 0.28528648614883423, "pix": 1.4146702289581299, "agree": 0.04928523302078247}
-{"step": 3950, "loss": 3.7998809814453125, "gnorm": 12.990020751953125, "s_per_step": 0.44319193127789075, "lat": 0.4436603784561157, "pix": 2.469829559326172, "agree": 0.08863910287618637}
-{"step": 4000, "loss": 5.267279624938965, "gnorm": 14.00953483581543, "s_per_step": 0.44309034472703934, "lat": 0.6878379583358765, "pix": 3.4613256454467773, "agree": 0.11181160807609558}
-{"val": {"psnr_r": 17.316714773575466, "psnr_f": 17.418252448240917, "lpips_r": 0.3291937040630728, "lpips_f": 0.3157626196431617, "sharp_r": 0.20774495438672602, "sharp_f": 0.6642827860390147, "agree": 0.1888549029827118, "struct": null, "app": null, "geo": null, "step": 4000}}

14_stream_fixer_s200_paired/train.log

@@ -1,400 +0,0 @@
-The config attributes {'clip_sample': False, 'sample_max_value': 1.0, 'set_alpha_to_one': False, 'skip_prk_steps': True} were passed to FlowMatchEulerDiscreteScheduler, but are not expected and will be ignored. Please verify your scheduler_config.json configuration file.
-The config attributes {'clip_sample': False, 'sample_max_value': 1.0, 'set_alpha_to_one': False, 'skip_prk_steps': True} were passed to FlowMatchEulerDiscreteScheduler, but are not expected and will be ignored. Please verify your scheduler_config.json configuration file.
-Cannot initialize model with low cpu memory usage because `accelerate` was not found in the environment. Defaulting to `low_cpu_mem_usage=False`. It is strongly recommended to install `accelerate` for faster and less memory-intense model loading. You can do so with: 
-```
-pip install accelerate
-```
-.
-The config attributes {'clip_sample': False, 'sample_max_value': 1.0, 'set_alpha_to_one': False, 'skip_prk_steps': True} were passed to FlowMatchEulerDiscreteScheduler, but are not expected and will be ignored. Please verify your scheduler_config.json configuration file.
-The config attributes {'clip_sample': False, 'sample_max_value': 1.0, 'set_alpha_to_one': False, 'skip_prk_steps': True} were passed to FlowMatchEulerDiscreteScheduler, but are not expected and will be ignored. Please verify your scheduler_config.json configuration file.
-Cannot initialize model with low cpu memory usage because `accelerate` was not found in the environment. Defaulting to `low_cpu_mem_usage=False`. It is strongly recommended to install `accelerate` for faster and less memory-intense model loading. You can do so with: 
-```
-pip install accelerate
-```
-.
-Setting up [LPIPS] perceptual loss: trunk [vgg], v[0.1], spatial [off]
-/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/site-packages/torchvision/models/_utils.py:208: UserWarning: The parameter 'pretrained' is deprecated since 0.13 and may be removed in the future, please use 'weights' instead.
-  warnings.warn(
-/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/site-packages/torchvision/models/_utils.py:223: UserWarning: Arguments other than a weight enum or `None` for 'weights' are deprecated since 0.13 and may be removed in the future. The current behavior is equivalent to passing `weights=VGG16_Weights.IMAGENET1K_V1`. You can also use `weights=VGG16_Weights.DEFAULT` to get the most up-to-date weights.
-  warnings.warn(msg)
-Setting up [LPIPS] perceptual loss: trunk [vgg], v[0.1], spatial [off]
-/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/site-packages/torchvision/models/_utils.py:208: UserWarning: The parameter 'pretrained' is deprecated since 0.13 and may be removed in the future, please use 'weights' instead.
-  warnings.warn(
-/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/site-packages/torchvision/models/_utils.py:223: UserWarning: Arguments other than a weight enum or `None` for 'weights' are deprecated since 0.13 and may be removed in the future. The current behavior is equivalent to passing `weights=VGG16_Weights.IMAGENET1K_V1`. You can also use `weights=VGG16_Weights.DEFAULT` to get the most up-to-date weights.
-  warnings.warn(msg)
-Loading model from: /mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/site-packages/lpips/weights/v0.1/vgg.pth
-Loading model from: /mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/site-packages/lpips/weights/v0.1/vgg.pth
-train windows 17214 (200 orbits), val windows 48, world 2
-VAL {"psnr_r": 17.3167, "psnr_f": 16.8381, "lpips_r": 0.3292, "lpips_f": 0.3579, "sharp_r": 0.2077, "sharp_f": 0.6178, "agree": 0.1955, "struct": null, "app": null, "geo": null, "step": 0}
-[rank1]:[W1102 20:04:01.766634832 reducer.cpp:1502] Warning: find_unused_parameters=True was specified in DDP constructor, but did not find any unused parameters in the forward pass. This flag results in an extra traversal of the autograd graph every iteration,  which can adversely affect performance. If your model indeed never has any unused parameters in the forward pass, consider turning this flag off. Note that this warning may be a false positive if your model has flow control causing later iterations to have unused parameters. (function operator())
-[rank0]:[W1102 20:04:01.810630774 reducer.cpp:1502] Warning: find_unused_parameters=True was specified in DDP constructor, but did not find any unused parameters in the forward pass. This flag results in an extra traversal of the autograd graph every iteration,  which can adversely affect performance. If your model indeed never has any unused parameters in the forward pass, consider turning this flag off. Note that this warning may be a false positive if your model has flow control causing later iterations to have unused parameters. (function operator())
-/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/free-latent-completion/scripts/syncfix/train_stream.py:250: UserWarning: Converting a tensor with requires_grad=True to a scalar may lead to unexpected behavior.
-Consider using tensor.detach() first. (Triggered internally at /pytorch/torch/csrc/autograd/generated/python_variable_methods.cpp:836.)
-  loss = lat + pix + a.w_agree * ag; info = dict(lat=float(lat), pix=float(pix), agree=float(ag))
-/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/free-latent-completion/scripts/syncfix/train_stream.py:250: UserWarning: Converting a tensor with requires_grad=True to a scalar may lead to unexpected behavior.
-Consider using tensor.detach() first. (Triggered internally at /pytorch/torch/csrc/autograd/generated/python_variable_methods.cpp:836.)
-  loss = lat + pix + a.w_agree * ag; info = dict(lat=float(lat), pix=float(pix), agree=float(ag))
-{"step": 50, "loss": 5.28296, "gnorm": 19.0123, "s_per_step": 0.46518, "lat": 0.34313, "pix": 3.69503, "agree": 0.12448}
-{"step": 100, "loss": 5.96665, "gnorm": 19.93128, "s_per_step": 0.45175, "lat": 0.63593, "pix": 3.60337, "agree": 0.17273}
-{"step": 150, "loss": 6.70642, "gnorm": 20.03252, "s_per_step": 0.44839, "lat": 0.41299, "pix": 4.52635, "agree": 0.17671}
-{"step": 200, "loss": 5.40998, "gnorm": 17.28683, "s_per_step": 0.44476, "lat": 0.52365, "pix": 3.48608, "agree": 0.14002}
-{"step": 250, "loss": 2.51725, "gnorm": 9.61781, "s_per_step": 0.44287, "lat": 0.12476, "pix": 1.9324, "agree": 0.04601}
-{"step": 300, "loss": 4.43017, "gnorm": 11.95686, "s_per_step": 0.44165, "lat": 0.35347, "pix": 2.96534, "agree": 0.11114}
-{"step": 350, "loss": 4.68809, "gnorm": 15.70925, "s_per_step": 0.44051, "lat": 0.37528, "pix": 3.08578, "agree": 0.1227}
-{"step": 400, "loss": 7.3148, "gnorm": 20.28812, "s_per_step": 0.43989, "lat": 0.66095, "pix": 4.75382, "agree": 0.19}
-{"step": 450, "loss": 2.88276, "gnorm": 7.82759, "s_per_step": 0.43892, "lat": 0.12323, "pix": 2.12297, "agree": 0.06366}
-{"step": 500, "loss": 5.69293, "gnorm": 11.12491, "s_per_step": 0.43826, "lat": 0.58423, "pix": 3.37119, "agree": 0.17375}
-VAL {"psnr_r": 17.3167, "psnr_f": 17.2562, "lpips_r": 0.3292, "lpips_f": 0.3278, "sharp_r": 0.2077, "sharp_f": 0.6296, "agree": 0.1943, "struct": null, "app": null, "geo": null, "step": 500}
-{"step": 550, "loss": 5.25703, "gnorm": 15.13127, "s_per_step": 0.44432, "lat": 0.43747, "pix": 3.66324, "agree": 0.11563}
-{"step": 600, "loss": 7.33764, "gnorm": 19.91977, "s_per_step": 0.44358, "lat": 0.67348, "pix": 5.05865, "agree": 0.16055}
-{"step": 650, "loss": 5.58404, "gnorm": 36.12888, "s_per_step": 0.44255, "lat": 0.40662, "pix": 3.67944, "agree": 0.1498}
-{"step": 700, "loss": 6.59393, "gnorm": 20.72497, "s_per_step": 0.44171, "lat": 0.58184, "pix": 4.30627, "agree": 0.17058}
-{"step": 750, "loss": 4.46368, "gnorm": 14.08114, "s_per_step": 0.44096, "lat": 0.34474, "pix": 3.11172, "agree": 0.10072}
-{"step": 800, "loss": 4.97641, "gnorm": 18.8637, "s_per_step": 0.44032, "lat": 0.7763, "pix": 3.11527, "agree": 0.10848}
-{"step": 850, "loss": 3.57932, "gnorm": 11.0161, "s_per_step": 0.43966, "lat": 0.63194, "pix": 2.42612, "agree": 0.05213}
-{"step": 900, "loss": 4.10833, "gnorm": 17.97794, "s_per_step": 0.43936, "lat": 0.32826, "pix": 2.82286, "agree": 0.09572}
-{"step": 950, "loss": 3.43479, "gnorm": 6.24893, "s_per_step": 0.43908, "lat": 0.33188, "pix": 2.25991, "agree": 0.0843}
-{"step": 1000, "loss": 3.45157, "gnorm": 9.02538, "s_per_step": 0.43865, "lat": 0.31045, "pix": 2.34786, "agree": 0.07933}
-VAL {"psnr_r": 17.3167, "psnr_f": 17.2838, "lpips_r": 0.3292, "lpips_f": 0.3257, "sharp_r": 0.2077, "sharp_f": 0.6613, "agree": 0.1919, "struct": null, "app": null, "geo": null, "step": 1000}
-{"step": 1050, "loss": 3.57158, "gnorm": 14.42652, "s_per_step": 0.44162, "lat": 0.38957, "pix": 2.04087, "agree": 0.11411}
-{"step": 1100, "loss": 4.5596, "gnorm": 8.70855, "s_per_step": 0.44111, "lat": 0.47088, "pix": 3.1535, "agree": 0.09352}
-{"step": 1150, "loss": 5.93217, "gnorm": 18.10099, "s_per_step": 0.44067, "lat": 0.47662, "pix": 3.86145, "agree": 0.15941}
-{"step": 1200, "loss": 2.79618, "gnorm": 9.36919, "s_per_step": 0.44051, "lat": 0.4849, "pix": 1.84972, "agree": 0.04616}
-{"step": 1250, "loss": 3.66321, "gnorm": 39.4276, "s_per_step": 0.44024, "lat": 0.31599, "pix": 2.69572, "agree": 0.06515}
-{"step": 1300, "loss": 2.7233, "gnorm": 8.12856, "s_per_step": 0.43991, "lat": 0.11198, "pix": 1.90663, "agree": 0.07047}
-{"step": 1350, "loss": 6.20073, "gnorm": 7.34594, "s_per_step": 0.43967, "lat": 0.58126, "pix": 4.50554, "agree": 0.11139}
-{"step": 1400, "loss": 1.67497, "gnorm": 29.47078, "s_per_step": 0.43954, "lat": 0.08111, "pix": 1.22086, "agree": 0.0373}
-{"step": 1450, "loss": 5.25488, "gnorm": 17.2422, "s_per_step": 0.43951, "lat": 0.67515, "pix": 3.39844, "agree": 0.11813}
-{"step": 1500, "loss": 2.30523, "gnorm": 7.86896, "s_per_step": 0.43941, "lat": 0.10623, "pix": 1.60007, "agree": 0.05989}
-VAL {"psnr_r": 17.3167, "psnr_f": 17.3415, "lpips_r": 0.3292, "lpips_f": 0.322, "sharp_r": 0.2077, "sharp_f": 0.6501, "agree": 0.1908, "struct": null, "app": null, "geo": null, "step": 1500}
-{"step": 1550, "loss": 4.35345, "gnorm": 12.09569, "s_per_step": 0.44154, "lat": 0.41743, "pix": 2.7483, "agree": 0.11877}
-{"step": 1600, "loss": 4.40761, "gnorm": 15.48571, "s_per_step": 0.44132, "lat": 0.41761, "pix": 3.13314, "agree": 0.08569}
-{"step": 1650, "loss": 4.7637, "gnorm": 15.48101, "s_per_step": 0.44119, "lat": 0.7621, "pix": 3.05472, "agree": 0.09469}
-{"step": 1700, "loss": 3.58298, "gnorm": 9.60507, "s_per_step": 0.44104, "lat": 0.40083, "pix": 2.38619, "agree": 0.0796}
-{"step": 1750, "loss": 2.6581, "gnorm": 26.26464, "s_per_step": 0.44087, "lat": 0.37014, "pix": 1.87566, "agree": 0.04123}
-{"step": 1800, "loss": 4.382, "gnorm": 10.63879, "s_per_step": 0.44075, "lat": 0.15205, "pix": 2.7893, "agree": 0.14407}
-{"step": 1850, "loss": 5.23642, "gnorm": 9.3723, "s_per_step": 0.44063, "lat": 0.33061, "pix": 3.7966, "agree": 0.11092}
-{"step": 1900, "loss": 3.84343, "gnorm": 12.53109, "s_per_step": 0.44045, "lat": 0.30072, "pix": 2.57348, "agree": 0.09692}
-{"step": 1950, "loss": 4.56698, "gnorm": 12.07399, "s_per_step": 0.44028, "lat": 0.46054, "pix": 3.21295, "agree": 0.08935}
-{"step": 2000, "loss": 2.8336, "gnorm": 7.93085, "s_per_step": 0.44019, "lat": 0.32638, "pix": 1.99943, "agree": 0.05078}
-VAL {"psnr_r": 17.3167, "psnr_f": 17.3456, "lpips_r": 0.3292, "lpips_f": 0.318, "sharp_r": 0.2077, "sharp_f": 0.6454, "agree": 0.192, "struct": null, "app": null, "geo": null, "step": 2000}
-{"step": 2050, "loss": 5.75942, "gnorm": 14.33886, "s_per_step": 0.44633, "lat": 0.43266, "pix": 3.84192, "agree": 0.14848}
-{"step": 2100, "loss": 3.20085, "gnorm": 6.44964, "s_per_step": 0.44618, "lat": 0.29197, "pix": 2.11645, "agree": 0.07924}
-{"step": 2150, "loss": 2.64712, "gnorm": 8.53653, "s_per_step": 0.44592, "lat": 0.08426, "pix": 1.96143, "agree": 0.06014}
-{"step": 2200, "loss": 2.80247, "gnorm": 5.26401, "s_per_step": 0.44569, "lat": 0.28327, "pix": 1.94547, "agree": 0.05737}
-{"step": 2250, "loss": 3.77119, "gnorm": 10.56261, "s_per_step": 0.44546, "lat": 0.10166, "pix": 2.79735, "agree": 0.08722}
-{"step": 2300, "loss": 5.90038, "gnorm": 32.10778, "s_per_step": 0.44518, "lat": 0.44576, "pix": 3.76156, "agree": 0.16931}
-{"step": 2350, "loss": 1.7644, "gnorm": 8.21197, "s_per_step": 0.44487, "lat": 0.21069, "pix": 1.19894, "agree": 0.03548}
-{"step": 2400, "loss": 2.54455, "gnorm": 11.13367, "s_per_step": 0.44466, "lat": 0.11066, "pix": 1.73279, "agree": 0.07011}
-{"step": 2450, "loss": 4.29611, "gnorm": 17.02275, "s_per_step": 0.44446, "lat": 0.61571, "pix": 2.90741, "agree": 0.0773}
-{"step": 2500, "loss": 3.71905, "gnorm": 29.36831, "s_per_step": 0.44422, "lat": 0.36689, "pix": 2.68537, "agree": 0.06668}
-VAL {"psnr_r": 17.3167, "psnr_f": 17.4168, "lpips_r": 0.3292, "lpips_f": 0.3186, "sharp_r": 0.2077, "sharp_f": 0.646, "agree": 0.1898, "struct": null, "app": null, "geo": null, "step": 2500}
-{"step": 2550, "loss": 3.22969, "gnorm": 13.6145, "s_per_step": 0.44536, "lat": 0.50995, "pix": 2.08665, "agree": 0.06331}
-{"step": 2600, "loss": 2.70281, "gnorm": 8.82827, "s_per_step": 0.44511, "lat": 0.30794, "pix": 1.93071, "agree": 0.04642}
-{"step": 2650, "loss": 2.9589, "gnorm": 7.46832, "s_per_step": 0.44493, "lat": 0.34192, "pix": 1.92939, "agree": 0.06876}
-{"step": 2700, "loss": 4.41385, "gnorm": 19.28908, "s_per_step": 0.44475, "lat": 0.32507, "pix": 3.1128, "agree": 0.0976}
-{"step": 2750, "loss": 5.24093, "gnorm": 33.16911, "s_per_step": 0.4446, "lat": 0.39316, "pix": 3.31263, "agree": 0.15351}
-{"step": 2800, "loss": 4.30253, "gnorm": 11.56947, "s_per_step": 0.4444, "lat": 0.15461, "pix": 2.7358, "agree": 0.14121}
-{"step": 2850, "loss": 3.62898, "gnorm": 11.84377, "s_per_step": 0.44425, "lat": 0.35865, "pix": 2.28747, "agree": 0.09829}
-{"step": 2900, "loss": 1.62814, "gnorm": 15.55265, "s_per_step": 0.44409, "lat": 0.07966, "pix": 1.07182, "agree": 0.04767}
-{"step": 2950, "loss": 3.45749, "gnorm": 7.93549, "s_per_step": 0.44396, "lat": 0.14037, "pix": 2.71582, "agree": 0.06013}
-{"step": 3000, "loss": 3.34258, "gnorm": 13.83503, "s_per_step": 0.44379, "lat": 0.53974, "pix": 2.01331, "agree": 0.07895}
-VAL {"psnr_r": 17.3167, "psnr_f": 17.3444, "lpips_r": 0.3292, "lpips_f": 0.3188, "sharp_r": 0.2077, "sharp_f": 0.7043, "agree": 0.1901, "struct": null, "app": null, "geo": null, "step": 3000}
-{"step": 3050, "loss": 5.01906, "gnorm": 15.61851, "s_per_step": 0.44479, "lat": 0.53513, "pix": 3.16676, "agree": 0.13172}
-{"step": 3100, "loss": 4.59153, "gnorm": 12.6261, "s_per_step": 0.44464, "lat": 0.34675, "pix": 3.24069, "agree": 0.10041}
-{"step": 3150, "loss": 2.71602, "gnorm": 11.15632, "s_per_step": 0.44447, "lat": 0.34542, "pix": 1.83167, "agree": 0.05389}
-{"step": 3200, "loss": 2.56267, "gnorm": 14.97436, "s_per_step": 0.44428, "lat": 0.30579, "pix": 1.79024, "agree": 0.04666}
-{"step": 3250, "loss": 3.24852, "gnorm": 56.41816, "s_per_step": 0.4442, "lat": 0.3222, "pix": 2.32841, "agree": 0.05979}
-{"step": 3300, "loss": 2.6899, "gnorm": 8.66203, "s_per_step": 0.44409, "lat": 0.32022, "pix": 1.81018, "agree": 0.05595}
-{"step": 3350, "loss": 2.35539, "gnorm": 8.35475, "s_per_step": 0.44394, "lat": 0.08787, "pix": 1.73051, "agree": 0.0537}
-{"step": 3400, "loss": 2.17085, "gnorm": 7.4371, "s_per_step": 0.44379, "lat": 0.12262, "pix": 1.71718, "agree": 0.0331}
-{"step": 3450, "loss": 5.30962, "gnorm": 12.23462, "s_per_step": 0.44364, "lat": 0.59083, "pix": 3.53778, "agree": 0.1181}
-{"step": 3500, "loss": 4.36844, "gnorm": 11.99641, "s_per_step": 0.44347, "lat": 0.42391, "pix": 2.87711, "agree": 0.10674}
-VAL {"psnr_r": 17.3167, "psnr_f": 17.3512, "lpips_r": 0.3292, "lpips_f": 0.3177, "sharp_r": 0.2077, "sharp_f": 0.7195, "agree": 0.1916, "struct": null, "app": null, "geo": null, "step": 3500}
-{"step": 3550, "loss": 3.25895, "gnorm": 7.00464, "s_per_step": 0.44434, "lat": 0.33431, "pix": 2.29281, "agree": 0.06318}
-{"step": 3600, "loss": 2.95003, "gnorm": 5.72332, "s_per_step": 0.44417, "lat": 0.14182, "pix": 2.01787, "agree": 0.07903}
-{"step": 3650, "loss": 2.96947, "gnorm": 6.42795, "s_per_step": 0.444, "lat": 0.31325, "pix": 1.97004, "agree": 0.06862}
-{"step": 3700, "loss": 2.94573, "gnorm": 15.64504, "s_per_step": 0.44382, "lat": 0.1236, "pix": 2.22591, "agree": 0.05962}
-{"step": 3750, "loss": 4.75014, "gnorm": 10.02662, "s_per_step": 0.44366, "lat": 0.37856, "pix": 3.26559, "agree": 0.1106}
-{"step": 3800, "loss": 5.50206, "gnorm": 15.83261, "s_per_step": 0.44351, "lat": 0.48081, "pix": 3.5999, "agree": 0.14213}
-{"step": 3850, "loss": 4.27548, "gnorm": 11.54232, "s_per_step": 0.44336, "lat": 0.30817, "pix": 3.07916, "agree": 0.08881}
-{"step": 3900, "loss": 2.19281, "gnorm": 8.54286, "s_per_step": 0.44328, "lat": 0.28529, "pix": 1.41467, "agree": 0.04929}
-{"step": 3950, "loss": 3.79988, "gnorm": 12.99002, "s_per_step": 0.44319, "lat": 0.44366, "pix": 2.46983, "agree": 0.08864}
-{"step": 4000, "loss": 5.26728, "gnorm": 14.00953, "s_per_step": 0.44309, "lat": 0.68784, "pix": 3.46133, "agree": 0.11181}
-VAL {"psnr_r": 17.3167, "psnr_f": 17.4183, "lpips_r": 0.3292, "lpips_f": 0.3158, "sharp_r": 0.2077, "sharp_f": 0.6643, "agree": 0.1889, "struct": null, "app": null, "geo": null, "step": 4000}
-Traceback (most recent call last):
-Traceback (most recent call last):
-Traceback (most recent call last):
-Traceback (most recent call last):
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 303, in _run_finalizers
-    finalizer()
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 227, in __call__
-    res = self._callback(*self._args, **self._kwargs)
-          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 136, in _remove_temp_dir
-    rmtree(tempdir, onerror=onerror)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 759, in rmtree
-    _rmtree_safe_fd(stack, onexc)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 303, in _run_finalizers
-    finalizer()
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 703, in _rmtree_safe_fd
-    onexc(func, path, err)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 750, in onexc
-    return onerror(func, path, exc_info)
-           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 227, in __call__
-    res = self._callback(*self._args, **self._kwargs)
-          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 700, in _rmtree_safe_fd
-    onexc(os.unlink, fullname, err)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 750, in onexc
-    return onerror(func, path, exc_info)
-           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 136, in _remove_temp_dir
-    rmtree(tempdir, onerror=onerror)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 698, in _rmtree_safe_fd
-    os.unlink(entry.name, dir_fd=topfd)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 759, in rmtree
-    _rmtree_safe_fd(stack, onexc)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 703, in _rmtree_safe_fd
-    onexc(func, path, err)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 303, in _run_finalizers
-    finalizer()
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 750, in onexc
-    return onerror(func, path, exc_info)
-           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 227, in __call__
-    res = self._callback(*self._args, **self._kwargs)
-          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 700, in _rmtree_safe_fd
-    onexc(os.unlink, fullname, err)
-OSError: [Errno 16] Device or resource busy: '/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/.tmp/pymp-krwhwa9v'
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 750, in onexc
-    return onerror(func, path, exc_info)
-           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 136, in _remove_temp_dir
-    rmtree(tempdir, onerror=onerror)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 303, in _run_finalizers
-    finalizer()
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 698, in _rmtree_safe_fd
-    os.unlink(entry.name, dir_fd=topfd)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 759, in rmtree
-    _rmtree_safe_fd(stack, onexc)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 703, in _rmtree_safe_fd
-    onexc(func, path, err)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 227, in __call__
-    res = self._callback(*self._args, **self._kwargs)
-          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 750, in onexc
-    return onerror(func, path, exc_info)
-           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 700, in _rmtree_safe_fd
-    onexc(os.unlink, fullname, err)
-OSError: [Errno 16] Device or resource busy: '/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/.tmp/pymp-7bcc42ps'
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 750, in onexc
-    return onerror(func, path, exc_info)
-           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 136, in _remove_temp_dir
-    rmtree(tempdir, onerror=onerror)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 698, in _rmtree_safe_fd
-    os.unlink(entry.name, dir_fd=topfd)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 759, in rmtree
-    _rmtree_safe_fd(stack, onexc)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 703, in _rmtree_safe_fd
-    onexc(func, path, err)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 750, in onexc
-    return onerror(func, path, exc_info)
-           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 700, in _rmtree_safe_fd
-    onexc(os.unlink, fullname, err)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 750, in onexc
-    return onerror(func, path, exc_info)
-           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-OSError: [Errno 16] Device or resource busy: '/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/.tmp/pymp-t7l4vf4v'
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 698, in _rmtree_safe_fd
-    os.unlink(entry.name, dir_fd=topfd)
-OSError: [Errno 16] Device or resource busy: '/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/.tmp/pymp-9r8cbfan'
-Traceback (most recent call last):
-Traceback (most recent call last):
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 303, in _run_finalizers
-    finalizer()
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 227, in __call__
-    res = self._callback(*self._args, **self._kwargs)
-          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 136, in _remove_temp_dir
-    rmtree(tempdir, onerror=onerror)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 759, in rmtree
-    _rmtree_safe_fd(stack, onexc)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 303, in _run_finalizers
-    finalizer()
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 703, in _rmtree_safe_fd
-    onexc(func, path, err)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 750, in onexc
-    return onerror(func, path, exc_info)
-           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 227, in __call__
-    res = self._callback(*self._args, **self._kwargs)
-          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 700, in _rmtree_safe_fd
-    onexc(os.unlink, fullname, err)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 136, in _remove_temp_dir
-    rmtree(tempdir, onerror=onerror)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 750, in onexc
-    return onerror(func, path, exc_info)
-           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 759, in rmtree
-    _rmtree_safe_fd(stack, onexc)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 698, in _rmtree_safe_fd
-    os.unlink(entry.name, dir_fd=topfd)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 703, in _rmtree_safe_fd
-    onexc(func, path, err)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 750, in onexc
-    return onerror(func, path, exc_info)
-           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 700, in _rmtree_safe_fd
-    onexc(os.unlink, fullname, err)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 750, in onexc
-    return onerror(func, path, exc_info)
-           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 698, in _rmtree_safe_fd
-    os.unlink(entry.name, dir_fd=topfd)
-OSError: [Errno 16] Device or resource busy: '/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/.tmp/pymp-5fg5psrg'
-OSError: [Errno 16] Device or resource busy: '/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/.tmp/pymp-6lw8vub8'
-Traceback (most recent call last):
-Traceback (most recent call last):
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 303, in _run_finalizers
-    finalizer()
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 227, in __call__
-    res = self._callback(*self._args, **self._kwargs)
-          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 136, in _remove_temp_dir
-    rmtree(tempdir, onerror=onerror)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 759, in rmtree
-    _rmtree_safe_fd(stack, onexc)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 703, in _rmtree_safe_fd
-    onexc(func, path, err)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 750, in onexc
-    return onerror(func, path, exc_info)
-           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 700, in _rmtree_safe_fd
-    onexc(os.unlink, fullname, err)
-Traceback (most recent call last):
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 750, in onexc
-    return onerror(func, path, exc_info)
-           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 698, in _rmtree_safe_fd
-    os.unlink(entry.name, dir_fd=topfd)
-OSError: [Errno 16] Device or resource busy: '/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/.tmp/pymp-xe815f19'
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 303, in _run_finalizers
-    finalizer()
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 227, in __call__
-    res = self._callback(*self._args, **self._kwargs)
-          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 136, in _remove_temp_dir
-    rmtree(tempdir, onerror=onerror)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 759, in rmtree
-    _rmtree_safe_fd(stack, onexc)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 703, in _rmtree_safe_fd
-    onexc(func, path, err)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 750, in onexc
-    return onerror(func, path, exc_info)
-           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 700, in _rmtree_safe_fd
-    onexc(os.unlink, fullname, err)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 750, in onexc
-    return onerror(func, path, exc_info)
-           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 698, in _rmtree_safe_fd
-    os.unlink(entry.name, dir_fd=topfd)
-OSError: [Errno 16] Device or resource busy: '/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/.tmp/pymp-bfn11tda'
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 303, in _run_finalizers
-    finalizer()
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 227, in __call__
-    res = self._callback(*self._args, **self._kwargs)
-          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 136, in _remove_temp_dir
-    rmtree(tempdir, onerror=onerror)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 759, in rmtree
-    _rmtree_safe_fd(stack, onexc)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 703, in _rmtree_safe_fd
-    onexc(func, path, err)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 750, in onexc
-    return onerror(func, path, exc_info)
-           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 700, in _rmtree_safe_fd
-    onexc(os.unlink, fullname, err)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 750, in onexc
-    return onerror(func, path, exc_info)
-           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 698, in _rmtree_safe_fd
-    os.unlink(entry.name, dir_fd=topfd)
-OSError: [Errno 16] Device or resource busy: '/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/.tmp/pymp-in4wnfc9'
-Traceback (most recent call last):
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 303, in _run_finalizers
-    finalizer()
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 227, in __call__
-    res = self._callback(*self._args, **self._kwargs)
-          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 136, in _remove_temp_dir
-    rmtree(tempdir, onerror=onerror)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 759, in rmtree
-    _rmtree_safe_fd(stack, onexc)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 703, in _rmtree_safe_fd
-    onexc(func, path, err)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 750, in onexc
-    return onerror(func, path, exc_info)
-           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 700, in _rmtree_safe_fd
-    onexc(os.unlink, fullname, err)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 750, in onexc
-    return onerror(func, path, exc_info)
-           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 698, in _rmtree_safe_fd
-    os.unlink(entry.name, dir_fd=topfd)
-OSError: [Errno 16] Device or resource busy: '/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/.tmp/pymp-2bfdgbtw'
-Traceback (most recent call last):
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 303, in _run_finalizers
-    finalizer()
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 227, in __call__
-    res = self._callback(*self._args, **self._kwargs)
-          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 136, in _remove_temp_dir
-    rmtree(tempdir, onerror=onerror)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 759, in rmtree
-    _rmtree_safe_fd(stack, onexc)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 703, in _rmtree_safe_fd
-    onexc(func, path, err)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 750, in onexc
-    return onerror(func, path, exc_info)
-           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 700, in _rmtree_safe_fd
-    onexc(os.unlink, fullname, err)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 750, in onexc
-    return onerror(func, path, exc_info)
-           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 698, in _rmtree_safe_fd
-    os.unlink(entry.name, dir_fd=topfd)
-OSError: [Errno 16] Device or resource busy: '/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/.tmp/pymp-u74yp5s7'
-Traceback (most recent call last):
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 303, in _run_finalizers
-    finalizer()
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 227, in __call__
-    res = self._callback(*self._args, **self._kwargs)
-          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/multiprocessing/util.py", line 136, in _remove_temp_dir
-    rmtree(tempdir, onerror=onerror)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 759, in rmtree
-    _rmtree_safe_fd(stack, onexc)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 703, in _rmtree_safe_fd
-    onexc(func, path, err)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 750, in onexc
-    return onerror(func, path, exc_info)
-           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 700, in _rmtree_safe_fd
-    onexc(os.unlink, fullname, err)
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 750, in onexc
-    return onerror(func, path, exc_info)
-           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-  File "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/miniconda3/envs/flc/lib/python3.12/shutil.py", line 698, in _rmtree_safe_fd
-    os.unlink(entry.name, dir_fd=topfd)
-OSError: [Errno 16] Device or resource busy: '/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/.tmp/pymp-494gz8qq'

15_output_fixers/README.md

@@ -1,25 +0,0 @@
-# 15_output_fixers
-
-Output fixers trained on 2026-11-03 (`scripts/syncfix/train_output.py`): they sharpen our memory's renders of a one-frame
-generation episode, with the real first frame as the reference.
-
-| file | restore to | size |
-|---|---|---|
-| `o0/denoiser_003000.pt` (+ `args.json`, `log.jsonl`) | `outputs/outfix/o0/` | 4.6 GB |
-| `o1/denoiser_003000.pt` (+ `args.json`, `log.jsonl`) | `outputs/outfix/o1/` | 4.6 GB |
-
-**Training.** Init from `14_stream_fixer_s200_paired/denoiser_004000.pt`; 3,000 steps, lr 1e-5, batch 2 x 4 GPUs. Pairs:
-`scripts/lyra/ours_alone.py --records` (our model's one-frame loop on 4K DL3DV training videos, memory rendered at 512 px;
-`outputs/outfix/pairs`, not uploaded). Losses: paired SyncFix step on frames whose 16 x 16 low-pass PSNR to the real frame is >= 18 dB,
-low-pass structure anchor to the render everywhere (weight 10). o1 adds the unpaired DINOv3 realism term (G2b, weight .5); o0 does not.
-
-**Use.**
-- o0: the headline "our model alone" arm on Lyra 2.0's Table 1 (comp_fix_of0; `scripts/lyra/outfix_videos.py`). RESULTS §0p:
-  DL3DV-Evaluation 55 LPIPS-Alex .511, T&T 5 .472 (Lyra 2.0 native .446 / .384).
-- o1: fixes our memory's renders before Lyra 2.0 sees them in the Lyra arm (`--ext-fix syncfix:outputs/outfix/o1/denoiser_003000.pt`;
-  RESULTS §0u, §0v), and the Lyra fine-tune's training conditions.
-- Held-out (16 videos): render sharpness 0.07 → 1.04 (o0) / 1.00 (o1) of real; aligned LPIPS .430 → .331 / .337. The realism term had
-  no measurable effect.
-
-**Terms.** Derived from SyncFix (no license file in github.com/Deming77/SyncFix); released on the project owner's decision
-(2026-11-04). Trained on DL3DV-derived frames (DL3DV-10K terms of use).

15_output_fixers/SHA256SUMS

@@ -1,7 +0,0 @@
-9950821ff4f3ca679be56809a65f5772b4629c8717da6dd66a3d23bccf7cacac  ./o0/args.json
-08a50937c6f91a0fbe83e14d7bfc6544f03388d4fb95f962327745924aa1bbfa  ./o0/denoiser_003000.pt
-7797438a3251d7859e269e8cf6b64bd7b0186ff17b7cf474c14625f332ebbdc8  ./o0/log.jsonl
-c7473198f4e5cf68237d996909163a5cad33fe81e750d9f9850ff1c334957c5a  ./o1/args.json
-ea2828f9e23e28e322f87f617ef2fc4f4cedcfb8850571c7c2c4c0877bb1c277  ./o1/denoiser_003000.pt
-43f654c611fb08a6d657912690b85640d364a6b1d336afa1aac633757134f12f  ./o1/log.jsonl
-fd78f832a1c56466c097ea5a57806ea3158748e731cc6a6620cf9f8168cfff86  ./README.md

15_output_fixers/o0/args.json

@@ -1,18 +0,0 @@
-{
- "out": "outputs/outfix/o0",
- "init": "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/free-latent-completion/outputs/syncfix_ft/s200_paired/denoiser_004000.pt",
- "align_db": "18.0",
- "w_pair": "1.0",
- "w_struct": "10.0",
- "w_app": "0.0",
- "realism_warmup": "500",
- "steps": "3000",
- "lr": "1e-05",
- "lr_warmup": "200",
- "batch": "2",
- "val_every": "500",
- "val_items": "64",
- "save_every": "1000",
- "workers": "4",
- "seed": "1103"
-}

15_output_fixers/o0/denoiser_003000.pt

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:08a50937c6f91a0fbe83e14d7bfc6544f03388d4fb95f962327745924aa1bbfa
-size 4853065399

15_output_fixers/o0/log.jsonl

@@ -1,67 +0,0 @@
-{"val": {"psnr_r": 18.29374998043745, "psnr_f": 18.03937694353935, "lpips_r": 0.42969857423733443, "lpips_f": 0.36465596770628905, "sharp_r": 0.0740657339648654, "sharp_f": 0.25356857387038567, "lowf_f": 30.119003542264302, "aligned": 0.325, "step": 0}}
-{"step": 50, "loss": 6.566572189331055, "gnorm": 174.52488708496094, "s_per_step": 1.13767493724823, "aligned": 0.5, "pair": 6.552228927612305, "struct": 0.01434314250946045, "app": 6.551360130310059, "w_app": 0.0}
-{"step": 100, "loss": 0.004289905074983835, "gnorm": 65.83573150634766, "s_per_step": 1.1993795442581177, "aligned": 0.0, "struct": 0.004289905074983835, "app": 6.314675807952881, "w_app": 0.0}
-{"step": 150, "loss": 3.604090929031372, "gnorm": 19.762821197509766, "s_per_step": 1.2746176179250082, "aligned": 0.5, "pair": 3.5912187099456787, "struct": 0.01287214644253254, "app": 6.741441249847412, "w_app": 0.0}
-{"step": 200, "loss": 0.014114519581198692, "gnorm": 24.476417541503906, "s_per_step": 1.3099539494514465, "aligned": 0.0, "struct": 0.014114519581198692, "app": 6.444731712341309, "w_app": 0.0}
-{"step": 250, "loss": 0.007112463936209679, "gnorm": 12.538662910461426, "s_per_step": 1.3355213718414307, "aligned": 0.0, "struct": 0.007112463936209679, "app": 6.596363067626953, "w_app": 0.0}
-{"step": 300, "loss": 0.004149497486650944, "gnorm": 2.7649760246276855, "s_per_step": 1.3343809660275776, "aligned": 0.0, "struct": 0.004149497486650944, "app": 6.6307878494262695, "w_app": 0.0}
-{"step": 350, "loss": 3.825540542602539, "gnorm": 9.637372016906738, "s_per_step": 1.313628932407924, "aligned": 0.5, "pair": 3.824531316757202, "struct": 0.0010092821903526783, "app": 6.695266246795654, "w_app": 0.0}
-{"step": 400, "loss": 4.969966888427734, "gnorm": 11.92798900604248, "s_per_step": 1.3076427727937698, "aligned": 0.5, "pair": 4.9609503746032715, "struct": 0.009016375988721848, "app": 6.267534255981445, "w_app": 0.0}
-{"step": 450, "loss": 2.398672103881836, "gnorm": 18.576133728027344, "s_per_step": 1.3106626749038697, "aligned": 0.5, "pair": 2.395437240600586, "struct": 0.0032347904052585363, "app": 6.272958278656006, "w_app": 0.0}
-{"step": 500, "loss": 0.005759264342486858, "gnorm": 45.00331115722656, "s_per_step": 1.3134973888397217, "aligned": 0.0, "struct": 0.005759264342486858, "app": 6.525854110717773, "w_app": 0.0}
-{"val": {"psnr_r": 18.29374998043745, "psnr_f": 18.19001887394832, "lpips_r": 0.42969857423733443, "lpips_f": 0.33644497509186083, "sharp_r": 0.0740657339648654, "sharp_f": 1.1614693095286688, "lowf_f": 32.61019020080566, "aligned": 0.325, "step": 500}}
-{"step": 550, "loss": 0.00464144442230463, "gnorm": 15.45079517364502, "s_per_step": 1.3268004742535677, "aligned": 0.0, "struct": 0.00464144442230463, "app": 6.638293743133545, "w_app": 0.0}
-{"step": 600, "loss": 0.0028280357364565134, "gnorm": 11.551639556884766, "s_per_step": 1.336089775164922, "aligned": 0.0, "struct": 0.0028280357364565134, "app": 6.4217634201049805, "w_app": 0.0}
-{"step": 650, "loss": 0.0020127957686781883, "gnorm": 9.634635925292969, "s_per_step": 1.33255487038539, "aligned": 0.0, "struct": 0.0020127957686781883, "app": 6.470822811126709, "w_app": 0.0}
-{"step": 700, "loss": 0.005437578074634075, "gnorm": 0.1506413221359253, "s_per_step": 1.3347669087137495, "aligned": 0.0, "struct": 0.005437578074634075, "app": 6.617572784423828, "w_app": 0.0}
-{"step": 750, "loss": 3.8120293617248535, "gnorm": 28.7796573638916, "s_per_step": 1.3356458549499513, "aligned": 0.5, "pair": 3.805140972137451, "struct": 0.006888285279273987, "app": 6.433078765869141, "w_app": 0.0}
-{"step": 800, "loss": 0.0024774973280727863, "gnorm": 0.1500473916530609, "s_per_step": 1.3350724557042122, "aligned": 0.0, "struct": 0.0024774973280727863, "app": 6.5122833251953125, "w_app": 0.0}
-{"step": 850, "loss": 0.004539792891591787, "gnorm": 21.699464797973633, "s_per_step": 1.3264304946450627, "aligned": 0.0, "struct": 0.004539792891591787, "app": 6.713287353515625, "w_app": 0.0}
-{"step": 900, "loss": 0.003328161546960473, "gnorm": 0.10590218007564545, "s_per_step": 1.327739777300093, "aligned": 0.0, "struct": 0.003328161546960473, "app": 6.442709922790527, "w_app": 0.0}
-{"step": 950, "loss": 0.0013097794726490974, "gnorm": 8.157008171081543, "s_per_step": 1.3282614369141428, "aligned": 0.0, "struct": 0.0013097794726490974, "app": 6.656916618347168, "w_app": 0.0}
-{"step": 1000, "loss": 6.843966007232666, "gnorm": 12.926344871520996, "s_per_step": 1.3286019165515899, "aligned": 0.5, "pair": 6.839400291442871, "struct": 0.0045656124129891396, "app": 6.515323638916016, "w_app": 0.0}
-{"val": {"psnr_r": 18.29374998043745, "psnr_f": 18.167191823323567, "lpips_r": 0.42969857423733443, "lpips_f": 0.3355032331668414, "sharp_r": 0.0740657339648654, "sharp_f": 1.3479778716961544, "lowf_f": 33.52021489143372, "aligned": 0.325, "step": 1000}}
-{"step": 1050, "loss": 0.0023787315003573895, "gnorm": 25.16949462890625, "s_per_step": 1.3423106877009074, "aligned": 0.0, "struct": 0.0023787315003573895, "app": 6.3543701171875, "w_app": 0.0}
-{"step": 1100, "loss": 3.40976619720459, "gnorm": 13.113037109375, "s_per_step": 1.341074687134136, "aligned": 1.0, "pair": 3.4041805267333984, "struct": 0.005585672799497843, "app": 6.63971471786499, "w_app": 0.0}
-{"step": 1150, "loss": 0.0059616342186927795, "gnorm": 25.532512664794922, "s_per_step": 1.3384263065586919, "aligned": 0.0, "struct": 0.0059616342186927795, "app": 6.335862636566162, "w_app": 0.0}
-{"step": 1200, "loss": 0.015089633874595165, "gnorm": 72.84829711914062, "s_per_step": 1.3364750309785207, "aligned": 0.0, "struct": 0.015089633874595165, "app": 6.703000545501709, "w_app": 0.0}
-{"step": 1250, "loss": 4.1286091804504395, "gnorm": 20.611492156982422, "s_per_step": 1.3410851831436157, "aligned": 0.5, "pair": 4.1184163093566895, "struct": 0.010192850604653358, "app": 6.701253890991211, "w_app": 0.0}
-{"step": 1300, "loss": 4.3941192626953125, "gnorm": 30.511133193969727, "s_per_step": 1.346707878112793, "aligned": 0.5, "pair": 4.391415596008301, "struct": 0.0027037542313337326, "app": 6.6548919677734375, "w_app": 0.0}
-{"step": 1350, "loss": 3.687105894088745, "gnorm": 57.65460205078125, "s_per_step": 1.349202143351237, "aligned": 0.5, "pair": 3.6828501224517822, "struct": 0.004255741368979216, "app": 6.569266319274902, "w_app": 0.0}
-{"step": 1400, "loss": 0.001917682122439146, "gnorm": 0.15982505679130554, "s_per_step": 1.3503045923369272, "aligned": 0.0, "struct": 0.001917682122439146, "app": 6.542843818664551, "w_app": 0.0}
-{"step": 1450, "loss": 0.0028349750209599733, "gnorm": 94.53026580810547, "s_per_step": 1.3479666797046004, "aligned": 0.0, "struct": 0.0028349750209599733, "app": 6.707538604736328, "w_app": 0.0}
-{"step": 1500, "loss": 0.0021214045118540525, "gnorm": 4.492269992828369, "s_per_step": 1.3503066515922546, "aligned": 0.0, "struct": 0.0021214045118540525, "app": 6.364469051361084, "w_app": 0.0}
-{"val": {"psnr_r": 18.29374998043745, "psnr_f": 18.291127987397022, "lpips_r": 0.42969857423733443, "lpips_f": 0.33608819926396394, "sharp_r": 0.0740657339648654, "sharp_f": 0.7950447777907054, "lowf_f": 34.9435632387797, "aligned": 0.325, "step": 1500}}
-{"step": 1550, "loss": 0.0045220134779810905, "gnorm": 31.964893341064453, "s_per_step": 1.349594789012786, "aligned": 0.0, "struct": 0.0045220134779810905, "app": 6.4993438720703125, "w_app": 0.0}
-{"step": 1600, "loss": 2.606477737426758, "gnorm": 11.12894058227539, "s_per_step": 1.3463747569918632, "aligned": 0.5, "pair": 2.5978586673736572, "struct": 0.008619073778390884, "app": 6.597753524780273, "w_app": 0.0}
-{"step": 1650, "loss": 2.5846776962280273, "gnorm": 8.57634162902832, "s_per_step": 1.3463826644781864, "aligned": 0.5, "pair": 2.581773519515991, "struct": 0.002904098015278578, "app": 6.5415568351745605, "w_app": 0.0}
-{"step": 1700, "loss": 0.0034744921140372753, "gnorm": 10.429474830627441, "s_per_step": 1.3490313785216388, "aligned": 0.0, "struct": 0.0034744921140372753, "app": 6.383647441864014, "w_app": 0.0}
-{"step": 1750, "loss": 8.631305694580078, "gnorm": 99.72422790527344, "s_per_step": 1.3532440560204642, "aligned": 0.5, "pair": 8.627215385437012, "struct": 0.004090378526598215, "app": 6.4338226318359375, "w_app": 0.0}
-{"step": 1800, "loss": 0.0023907648865133524, "gnorm": 4.270789623260498, "s_per_step": 1.3557165267732407, "aligned": 0.0, "struct": 0.0023907648865133524, "app": 6.655078887939453, "w_app": 0.0}
-{"step": 1850, "loss": 0.007275600917637348, "gnorm": 14.193973541259766, "s_per_step": 1.3578658480257602, "aligned": 0.0, "struct": 0.007275600917637348, "app": 6.460803031921387, "w_app": 0.0}
-{"step": 1900, "loss": 5.193690776824951, "gnorm": 78.59044647216797, "s_per_step": 1.3595412394874975, "aligned": 1.0, "pair": 5.185530185699463, "struct": 0.008160468190908432, "app": 7.311769008636475, "w_app": 0.0}
-{"step": 1950, "loss": 6.91905403137207, "gnorm": 10.173983573913574, "s_per_step": 1.3597608904960827, "aligned": 0.5, "pair": 6.916546821594238, "struct": 0.0025071369018405676, "app": 6.537930965423584, "w_app": 0.0}
-{"step": 2000, "loss": 0.0023671784438192844, "gnorm": 40.09288787841797, "s_per_step": 1.3585990983247758, "aligned": 0.0, "struct": 0.0023671784438192844, "app": 6.4348297119140625, "w_app": 0.0}
-{"val": {"psnr_r": 18.29374998043745, "psnr_f": 18.223347468253895, "lpips_r": 0.42969857423733443, "lpips_f": 0.33390097663952756, "sharp_r": 0.0740657339648654, "sharp_f": 0.9449760509033998, "lowf_f": 35.69722615877787, "aligned": 0.325, "step": 2000}}
-{"step": 2050, "loss": 0.0033338633365929127, "gnorm": 30.524206161499023, "s_per_step": 1.3652645619322614, "aligned": 0.0, "struct": 0.0033338633365929127, "app": 6.608201026916504, "w_app": 0.0}
-{"step": 2100, "loss": 3.0987887382507324, "gnorm": 26.457218170166016, "s_per_step": 1.3653414011001588, "aligned": 0.5, "pair": 3.092071294784546, "struct": 0.006717368494719267, "app": 6.529991149902344, "w_app": 0.0}
-{"step": 2150, "loss": 0.0025633517652750015, "gnorm": 51.26750183105469, "s_per_step": 1.3668240507258926, "aligned": 0.0, "struct": 0.0025633517652750015, "app": 6.669938087463379, "w_app": 0.0}
-{"step": 2200, "loss": 2.2181153297424316, "gnorm": 12.567102432250977, "s_per_step": 1.3664327843622728, "aligned": 0.5, "pair": 2.2152645587921143, "struct": 0.0028506594244390726, "app": 6.499617576599121, "w_app": 0.0}
-{"step": 2250, "loss": 0.0023828591220080853, "gnorm": 47.96087646484375, "s_per_step": 1.3664780999289619, "aligned": 0.0, "struct": 0.0023828591220080853, "app": 6.8202338218688965, "w_app": 0.0}
-{"step": 2300, "loss": 0.0026550099719315767, "gnorm": 6.0671467781066895, "s_per_step": 1.3682480787194293, "aligned": 0.0, "struct": 0.0026550099719315767, "app": 6.590298652648926, "w_app": 0.0}
-{"step": 2350, "loss": 4.079659938812256, "gnorm": 30.678058624267578, "s_per_step": 1.3684940717575398, "aligned": 0.5, "pair": 4.073484420776367, "struct": 0.0061753783375024796, "app": 6.688738822937012, "w_app": 0.0}
-{"step": 2400, "loss": 0.00147032190579921, "gnorm": 8.121889114379883, "s_per_step": 1.3667073467373847, "aligned": 0.0, "struct": 0.00147032190579921, "app": 6.5845489501953125, "w_app": 0.0}
-{"step": 2450, "loss": 3.790067672729492, "gnorm": 17.827836990356445, "s_per_step": 1.3681556087610673, "aligned": 0.5, "pair": 3.7884702682495117, "struct": 0.001597365946508944, "app": 6.74163818359375, "w_app": 0.0}
-{"step": 2500, "loss": 0.00202089361846447, "gnorm": 5.223774433135986, "s_per_step": 1.3693249008178712, "aligned": 0.0, "struct": 0.00202089361846447, "app": 6.539862632751465, "w_app": 0.0}
-{"val": {"psnr_r": 18.29374998043745, "psnr_f": 18.08383264297094, "lpips_r": 0.42969857423733443, "lpips_f": 0.33555016265465665, "sharp_r": 0.0740657339648654, "sharp_f": 0.9656349629163742, "lowf_f": 35.678167247772215, "aligned": 0.325, "step": 2500}}
-{"step": 2550, "loss": 2.2987310886383057, "gnorm": 21.874475479125977, "s_per_step": 1.3713712334165387, "aligned": 0.5, "pair": 2.2944676876068115, "struct": 0.004263500217348337, "app": 7.2176923751831055, "w_app": 0.0}
-{"step": 2600, "loss": 11.673836708068848, "gnorm": 358.60626220703125, "s_per_step": 1.372596648473006, "aligned": 0.5, "pair": 11.670365333557129, "struct": 0.0034713433124125004, "app": 6.36972713470459, "w_app": 0.0}
-{"step": 2650, "loss": 0.0024286017287522554, "gnorm": 81.49159240722656, "s_per_step": 1.3742716718169878, "aligned": 0.0, "struct": 0.0024286017287522554, "app": 6.370230674743652, "w_app": 0.0}
-{"step": 2700, "loss": 0.004091513808816671, "gnorm": 0.08713878691196442, "s_per_step": 1.3760021523193078, "aligned": 0.0, "struct": 0.004091513808816671, "app": 6.572257041931152, "w_app": 0.0}
-{"step": 2750, "loss": 0.0046708183363080025, "gnorm": 8.568123817443848, "s_per_step": 1.376009539084001, "aligned": 0.0, "struct": 0.0046708183363080025, "app": 6.487266540527344, "w_app": 0.0}
-{"step": 2800, "loss": 0.0019057339522987604, "gnorm": 35.28984069824219, "s_per_step": 1.376651193840163, "aligned": 0.0, "struct": 0.0019057339522987604, "app": 6.702491760253906, "w_app": 0.0}
-{"step": 2850, "loss": 0.0014136284589767456, "gnorm": 51.944488525390625, "s_per_step": 1.376607826467146, "aligned": 0.0, "struct": 0.0014136284589767456, "app": 6.670121192932129, "w_app": 0.0}
-{"step": 2900, "loss": 6.0533881187438965, "gnorm": 39.868465423583984, "s_per_step": 1.3754044644585972, "aligned": 0.5, "pair": 6.048434734344482, "struct": 0.004953497555106878, "app": 6.668424129486084, "w_app": 0.0}
-{"step": 2950, "loss": 0.009781298227608204, "gnorm": 38.471031188964844, "s_per_step": 1.3737625088125973, "aligned": 0.0, "struct": 0.009781298227608204, "app": 6.650856018066406, "w_app": 0.0}
-{"step": 3000, "loss": 0.006936336867511272, "gnorm": 22.076435089111328, "s_per_step": 1.3730210908253988, "aligned": 0.0, "struct": 0.006936336867511272, "app": 6.52136754989624, "w_app": 0.0}
-{"val": {"psnr_r": 18.29374998043745, "psnr_f": 18.282010763119427, "lpips_r": 0.42969857423733443, "lpips_f": 0.3314247379700343, "sharp_r": 0.0740657339648654, "sharp_f": 1.03555294200778, "lowf_f": 35.7269850730896, "aligned": 0.325, "step": 3000}}

15_output_fixers/o1/args.json

@@ -1,18 +0,0 @@
-{
- "out": "outputs/outfix/o1",
- "init": "/mnt/FileSystem-20261019-1231-05/tianmin/yyin34/free-latent-completion/outputs/syncfix_ft/s200_paired/denoiser_004000.pt",
- "align_db": "18.0",
- "w_pair": "1.0",
- "w_struct": "10.0",
- "w_app": "0.5",
- "realism_warmup": "500",
- "steps": "3000",
- "lr": "1e-05",
- "lr_warmup": "200",
- "batch": "2",
- "val_every": "500",
- "val_items": "64",
- "save_every": "1000",
- "workers": "4",
- "seed": "1103"
-}

15_output_fixers/o1/denoiser_003000.pt

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:ea2828f9e23e28e322f87f617ef2fc4f4cedcfb8850571c7c2c4c0877bb1c277
-size 4853065399

15_output_fixers/o1/log.jsonl

@@ -1,67 +0,0 @@
-{"val": {"psnr_r": 18.29374998043745, "psnr_f": 18.03937694353935, "lpips_r": 0.42969857423733443, "lpips_f": 0.36465596770628905, "sharp_r": 0.0740657339648654, "sharp_f": 0.25356857387038567, "lowf_f": 30.119003542264302, "aligned": 0.325, "step": 0}}
-{"step": 50, "loss": 6.552639007568359, "gnorm": 177.91310119628906, "s_per_step": 1.220733437538147, "aligned": 0.5, "pair": 6.537889003753662, "struct": 0.0143181923776865, "app": 6.552777290344238, "w_app": 6.595887825824321e-05}
-{"step": 100, "loss": 0.004610520321875811, "gnorm": 68.49533081054688, "s_per_step": 1.2613369011878968, "aligned": 0.0, "struct": 0.004201072733849287, "app": 6.279748439788818, "w_app": 6.520127062685788e-05}
-{"step": 150, "loss": 3.6091973781585693, "gnorm": 19.210718154907227, "s_per_step": 1.3069580221176147, "aligned": 0.5, "pair": 3.593662738800049, "struct": 0.014077942818403244, "app": 6.749533653259277, "w_app": 0.0002158443268854171}
-{"step": 200, "loss": 0.016183845698833466, "gnorm": 30.35321807861328, "s_per_step": 1.3254887199401855, "aligned": 0.0, "struct": 0.014941115863621235, "app": 6.468802452087402, "w_app": 0.0001921111484989524}
-{"step": 250, "loss": 0.009517651982605457, "gnorm": 12.47537899017334, "s_per_step": 1.330306923866272, "aligned": 0.0, "struct": 0.008383899927139282, "app": 6.633281707763672, "w_app": 0.00017091876361519098}
-{"step": 300, "loss": 0.00444971164688468, "gnorm": 2.644144296646118, "s_per_step": 1.3304858342806498, "aligned": 0.0, "struct": 0.003487352514639497, "app": 6.668112754821777, "w_app": 0.00014432254247367382}
-{"step": 350, "loss": 3.9645488262176514, "gnorm": 12.778855323791504, "s_per_step": 1.3189399439947946, "aligned": 0.5, "pair": 3.9628334045410156, "struct": 0.0010598166845738888, "app": 6.654792785644531, "w_app": 9.852494549704715e-05}
-{"step": 400, "loss": 4.825254917144775, "gnorm": 16.242292404174805, "s_per_step": 1.3066473031044006, "aligned": 0.5, "pair": 4.813778400421143, "struct": 0.009721125476062298, "app": 6.309676170349121, "w_app": 0.00027814859058707955}
-{"step": 450, "loss": 2.3680508136749268, "gnorm": 14.772153854370117, "s_per_step": 1.3159028869205052, "aligned": 0.5, "pair": 2.3642752170562744, "struct": 0.002991965040564537, "app": 6.271307945251465, "w_app": 0.00012497743737185373}
-{"step": 500, "loss": 0.007379099726676941, "gnorm": 21.214921951293945, "s_per_step": 1.319250349998474, "aligned": 0.0, "struct": 0.004701812285929918, "app": 6.387717247009277, "w_app": 0.00041913051973097026}
-{"val": {"psnr_r": 18.29374998043745, "psnr_f": 18.078568360744377, "lpips_r": 0.42969857423733443, "lpips_f": 0.33643297163339764, "sharp_r": 0.0740657339648654, "sharp_f": 1.0677654035389423, "lowf_f": 33.40557478268941, "aligned": 0.325, "step": 500}}
-{"step": 550, "loss": 0.008051928132772446, "gnorm": 15.199132919311523, "s_per_step": 1.3345215554670855, "aligned": 0.0, "struct": 0.006302442401647568, "app": 6.606564998626709, "w_app": 0.00026481019449420273}
-{"step": 600, "loss": 0.004130398854613304, "gnorm": 18.616392135620117, "s_per_step": 1.336434987783432, "aligned": 0.0, "struct": 0.0028501837514340878, "app": 6.490939140319824, "w_app": 0.00019723112927749753}
-{"step": 650, "loss": 0.003574131056666374, "gnorm": 16.77350425720215, "s_per_step": 1.3377108771984394, "aligned": 0.0, "struct": 0.0024263528175652027, "app": 6.356172561645508, "w_app": 0.0001805769425118342}
-{"step": 700, "loss": 0.008294477127492428, "gnorm": 0.22166825830936432, "s_per_step": 1.3377707304273332, "aligned": 0.0, "struct": 0.00626399228349328, "app": 6.6199188232421875, "w_app": 0.00030672355205751956}
-{"step": 750, "loss": 3.714414358139038, "gnorm": 36.63033676147461, "s_per_step": 1.3278345114390055, "aligned": 0.5, "pair": 3.7052717208862305, "struct": 0.006646262481808662, "app": 6.4053754806518555, "w_app": 0.0003897368151228875}
-{"step": 800, "loss": 0.004603363107889891, "gnorm": 0.1637822389602661, "s_per_step": 1.3243983089923859, "aligned": 0.0, "struct": 0.0028517572209239006, "app": 6.375239849090576, "w_app": 0.00027475139359012246}
-{"step": 850, "loss": 0.0068963211961090565, "gnorm": 16.175033569335938, "s_per_step": 1.3254515286052928, "aligned": 0.0, "struct": 0.0051343184895813465, "app": 6.699587821960449, "w_app": 0.0002630016824696213}
-{"step": 900, "loss": 0.004637528210878372, "gnorm": 0.0848270058631897, "s_per_step": 1.3273523733350965, "aligned": 0.0, "struct": 0.003958428744226694, "app": 6.597204208374023, "w_app": 0.00010293749073753133}
-{"step": 950, "loss": 0.0028068211395293474, "gnorm": 12.68388557434082, "s_per_step": 1.3263090186370046, "aligned": 0.0, "struct": 0.0015652799047529697, "app": 6.575273036956787, "w_app": 0.0001888197148218751}
-{"step": 1000, "loss": 7.023467063903809, "gnorm": 22.829139709472656, "s_per_step": 1.3298970222473145, "aligned": 0.5, "pair": 7.016800880432129, "struct": 0.004750752821564674, "app": 6.590871334075928, "w_app": 0.0002906487789005041}
-{"val": {"psnr_r": 18.29374998043745, "psnr_f": 18.138446832314514, "lpips_r": 0.42969857423733443, "lpips_f": 0.33579823756829286, "sharp_r": 0.0740657339648654, "sharp_f": 1.2918829798698426, "lowf_f": 33.842165295283, "aligned": 0.325, "step": 1000}}
-{"step": 1050, "loss": 0.0053086550906300545, "gnorm": 26.678897857666016, "s_per_step": 1.3489450382051014, "aligned": 0.0, "struct": 0.0032122228294610977, "app": 6.314403533935547, "w_app": 0.0003320079995319247}
-{"step": 1100, "loss": 3.431063652038574, "gnorm": 21.39289093017578, "s_per_step": 1.3480983146754177, "aligned": 1.0, "pair": 3.4227182865142822, "struct": 0.005218227859586477, "app": 6.697114944458008, "w_app": 0.0004669180198106915}
-{"step": 1150, "loss": 0.009197553619742393, "gnorm": 19.968372344970703, "s_per_step": 1.3458918034512064, "aligned": 0.0, "struct": 0.006401968188583851, "app": 6.305885314941406, "w_app": 0.00044332953984849155}
-{"step": 1200, "loss": 0.019687265157699585, "gnorm": 55.18376159667969, "s_per_step": 1.3440483742952347, "aligned": 0.0, "struct": 0.0165328960865736, "app": 6.592324733734131, "w_app": 0.000478491245303303}
-{"step": 1250, "loss": 4.1878862380981445, "gnorm": 24.727344512939453, "s_per_step": 1.3437046312332153, "aligned": 0.5, "pair": 4.174732208251953, "struct": 0.009137708693742752, "app": 6.663069725036621, "w_app": 0.0006028150673955679}
-{"step": 1300, "loss": 4.255566596984863, "gnorm": 25.522991180419922, "s_per_step": 1.3454684448242187, "aligned": 0.5, "pair": 4.2502665519714355, "struct": 0.0032641785219311714, "app": 6.647686004638672, "w_app": 0.00030629505636170506}
-{"step": 1350, "loss": 3.7268385887145996, "gnorm": 70.3050308227539, "s_per_step": 1.345767090232284, "aligned": 0.5, "pair": 3.7193846702575684, "struct": 0.005747064016759396, "app": 6.582553863525391, "w_app": 0.00025929402909241617}
-{"step": 1400, "loss": 0.003252290887758136, "gnorm": 0.28614741563796997, "s_per_step": 1.344504028899329, "aligned": 0.0, "struct": 0.0018791256006807089, "app": 6.4920573234558105, "w_app": 0.0002115146635333076}
-{"step": 1450, "loss": 0.0050394292920827866, "gnorm": 95.5965347290039, "s_per_step": 1.339054954463038, "aligned": 0.0, "struct": 0.002751614898443222, "app": 6.730399131774902, "w_app": 0.00033992258249782026}
-{"step": 1500, "loss": 0.0037938207387924194, "gnorm": 3.7528035640716553, "s_per_step": 1.3334797937075298, "aligned": 0.0, "struct": 0.0020684022456407547, "app": 6.302001953125, "w_app": 0.0002737889299169183}
-{"val": {"psnr_r": 18.29374998043745, "psnr_f": 18.282181690900753, "lpips_r": 0.42969857423733443, "lpips_f": 0.33553054966987705, "sharp_r": 0.0740657339648654, "sharp_f": 0.7763521226743857, "lowf_f": 35.54996166229248, "aligned": 0.325, "step": 1500}}
-{"step": 1550, "loss": 0.006193590350449085, "gnorm": 32.25614547729492, "s_per_step": 1.336506017254245, "aligned": 0.0, "struct": 0.00440203957259655, "app": 6.449469566345215, "w_app": 0.0002777826739475131}
-{"step": 1600, "loss": 2.450453042984009, "gnorm": 5.20603609085083, "s_per_step": 1.3372031006217002, "aligned": 0.5, "pair": 2.442251682281494, "struct": 0.006715167313814163, "app": 6.642580986022949, "w_app": 0.00022374212858267128}
-{"step": 1650, "loss": 2.745610475540161, "gnorm": 11.657225608825684, "s_per_step": 1.3365183801362008, "aligned": 0.5, "pair": 2.7393949031829834, "struct": 0.004166564904153347, "app": 6.5819315910339355, "w_app": 0.00031130839488469064}
-{"step": 1700, "loss": 0.0058858348056674, "gnorm": 10.615862846374512, "s_per_step": 1.3390158268984627, "aligned": 0.0, "struct": 0.0039636180736124516, "app": 6.421577453613281, "w_app": 0.0002993371454067528}
-{"step": 1750, "loss": 4.893038272857666, "gnorm": 39.39244842529297, "s_per_step": 1.3413186592374529, "aligned": 0.5, "pair": 4.887107849121094, "struct": 0.004359228070825338, "app": 6.464790344238281, "w_app": 0.00024303642567247152}
-{"step": 1800, "loss": 0.004941133316606283, "gnorm": 4.468908786773682, "s_per_step": 1.3424256428082784, "aligned": 0.0, "struct": 0.003428701777011156, "app": 6.680632591247559, "w_app": 0.00022639046073891222}
-{"step": 1850, "loss": 0.010022862814366817, "gnorm": 14.476775169372559, "s_per_step": 1.3437472829303225, "aligned": 0.0, "struct": 0.007346238940954208, "app": 6.538488388061523, "w_app": 0.0004093643219675869}
-{"step": 1900, "loss": 4.9621405601501465, "gnorm": 61.03131866455078, "s_per_step": 1.343356147063406, "aligned": 1.0, "pair": 4.95025634765625, "struct": 0.008230739273130894, "app": 7.280120849609375, "w_app": 0.0005018756492063403}
-{"step": 1950, "loss": 7.059752464294434, "gnorm": 16.561548233032227, "s_per_step": 1.3432337450369811, "aligned": 0.5, "pair": 7.055301666259766, "struct": 0.0026594779919832945, "app": 6.567132949829102, "w_app": 0.0002727908722590655}
-{"step": 2000, "loss": 0.005557009018957615, "gnorm": 43.20547866821289, "s_per_step": 1.3425990988016128, "aligned": 0.0, "struct": 0.002989082131534815, "app": 6.404223918914795, "w_app": 0.00040097394958138466}
-{"val": {"psnr_r": 18.29374998043745, "psnr_f": 18.16419709034455, "lpips_r": 0.42969857423733443, "lpips_f": 0.33674179781705904, "sharp_r": 0.0740657339648654, "sharp_f": 0.9628280060986678, "lowf_f": 35.8948610941569, "aligned": 0.325, "step": 2000}}
-{"step": 2050, "loss": 0.005294512026011944, "gnorm": 23.749683380126953, "s_per_step": 1.3476777945495233, "aligned": 0.0, "struct": 0.0032754007261246443, "app": 6.635986328125, "w_app": 0.00030426698504015803}
-{"step": 2100, "loss": 3.1800475120544434, "gnorm": 26.210491180419922, "s_per_step": 1.346140885466621, "aligned": 0.5, "pair": 3.171590566635132, "struct": 0.006976816803216934, "app": 6.606240749359131, "w_app": 0.00022405855997931212}
-{"step": 2150, "loss": 0.0055535826832056046, "gnorm": 26.91501808166504, "s_per_step": 1.343878425997357, "aligned": 0.0, "struct": 0.0036286553367972374, "app": 6.749362468719482, "w_app": 0.0002852013858500868}
-{"step": 2200, "loss": 2.199176073074341, "gnorm": 9.567594528198242, "s_per_step": 1.3438086666844107, "aligned": 0.5, "pair": 2.1928091049194336, "struct": 0.004111044574528933, "app": 6.490529537200928, "w_app": 0.00034755223896354437}
-{"step": 2250, "loss": 0.004574543330818415, "gnorm": 41.73478317260742, "s_per_step": 1.344584704928928, "aligned": 0.0, "struct": 0.0025169923901557922, "app": 6.714572906494141, "w_app": 0.000306430651107803}
-{"step": 2300, "loss": 0.005329352803528309, "gnorm": 5.598484039306641, "s_per_step": 1.343384502970654, "aligned": 0.0, "struct": 0.00305784959346056, "app": 6.5654449462890625, "w_app": 0.0003459785657469183}
-{"step": 2350, "loss": 4.0723490715026855, "gnorm": 28.414316177368164, "s_per_step": 1.3439243955815092, "aligned": 0.5, "pair": 4.065491199493408, "struct": 0.005578360520303249, "app": 6.727513790130615, "w_app": 0.00019019958563148975}
-{"step": 2400, "loss": 0.0024675317108631134, "gnorm": 8.207681655883789, "s_per_step": 1.3426325523853302, "aligned": 0.0, "struct": 0.0012457272969186306, "app": 6.577241897583008, "w_app": 0.00018576243019197136}
-{"step": 2450, "loss": 3.718296766281128, "gnorm": 41.282737731933594, "s_per_step": 1.3418749485210497, "aligned": 0.5, "pair": 3.715717077255249, "struct": 0.001934513682499528, "app": 6.724120140075684, "w_app": 9.594007133273408e-05}
-{"step": 2500, "loss": 0.0035813089925795794, "gnorm": 6.943805694580078, "s_per_step": 1.3387886953353882, "aligned": 0.0, "struct": 0.002643969841301441, "app": 6.586852073669434, "w_app": 0.00014230457600206137}
-{"val": {"psnr_r": 18.29374998043745, "psnr_f": 18.222498478033604, "lpips_r": 0.42969857423733443, "lpips_f": 0.3320395400126775, "sharp_r": 0.0740657339648654, "sharp_f": 1.0352707174917062, "lowf_f": 35.46252930959066, "aligned": 0.325, "step": 2500}}
-{"step": 2550, "loss": 2.3472208976745605, "gnorm": 22.362241744995117, "s_per_step": 1.3393541925093706, "aligned": 0.5, "pair": 2.3435916900634766, "struct": 0.0030962456949055195, "app": 7.112529754638672, "w_app": 7.492117583751678e-05}
-{"step": 2600, "loss": 10.028396606445312, "gnorm": 242.53533935546875, "s_per_step": 1.3382644894489875, "aligned": 0.5, "pair": 10.023185729980469, "struct": 0.003439165884628892, "app": 6.351202011108398, "w_app": 0.00027894365484826267}
-{"step": 2650, "loss": 0.00286307604983449, "gnorm": 84.79878234863281, "s_per_step": 1.338910433661263, "aligned": 0.0, "struct": 0.0020848081912845373, "app": 6.408143520355225, "w_app": 0.00012144983338657767}
-{"step": 2700, "loss": 0.008468177169561386, "gnorm": 0.09797988086938858, "s_per_step": 1.3391944020765798, "aligned": 0.0, "struct": 0.006479677278548479, "app": 6.611115455627441, "w_app": 0.0003007812483701855}
-{"step": 2750, "loss": 0.005553849041461945, "gnorm": 10.018019676208496, "s_per_step": 1.338002706267617, "aligned": 0.0, "struct": 0.003879243042320013, "app": 6.5763678550720215, "w_app": 0.0002546399482525885}
-{"step": 2800, "loss": 0.0030699013732373714, "gnorm": 27.493104934692383, "s_per_step": 1.336248282619885, "aligned": 0.0, "struct": 0.002369778463616967, "app": 6.661778450012207, "w_app": 0.00010509548883419484}
-{"step": 2850, "loss": 0.002022217959165573, "gnorm": 29.878677368164062, "s_per_step": 1.3359079125889561, "aligned": 0.0, "struct": 0.0012107781367376447, "app": 6.646629333496094, "w_app": 0.00012208291445858777}
-{"step": 2900, "loss": 5.757727146148682, "gnorm": 46.693084716796875, "s_per_step": 1.3350471757198201, "aligned": 0.5, "pair": 5.751986503601074, "struct": 0.003599480725824833, "app": 6.641255855560303, "w_app": 0.00032239448046311736}
-{"step": 2950, "loss": 0.01205320842564106, "gnorm": 42.87821960449219, "s_per_step": 1.3346032212144237, "aligned": 0.0, "struct": 0.010720577090978622, "app": 6.573702812194824, "w_app": 0.00020272158144507557}
-{"step": 3000, "loss": 0.010251899249851704, "gnorm": 28.192493438720703, "s_per_step": 1.3346000096003214, "aligned": 0.0, "struct": 0.008650366216897964, "app": 6.491092681884766, "w_app": 0.0002467277809046209}
-{"val": {"psnr_r": 18.29374998043745, "psnr_f": 18.064937689365486, "lpips_r": 0.42969857423733443, "lpips_f": 0.33663131640507626, "sharp_r": 0.0740657339648654, "sharp_f": 0.9986916191875934, "lowf_f": 35.076684141159056, "aligned": 0.325, "step": 3000}}

16_gen3r_round2_probe/README.md

@@ -1,39 +0,0 @@
-# 16_gen3r_round2_probe
-
-Everything the Gen3R round-2 probe produced (2026-11-03; plan docs/model/GEN3R_ROUND2_PLAN_2026-11-03.md stage 0; results
-docs/RESULTS.md §0q "Round-2 probe"). With these files the probe can be re-scored, re-analysed or extended (new variants on the same
-memory states) without rebuilding the memories or rerunning Gen3R.
-
-| file | restore | size |
-|---|---|---|
-| `r2probe.zip` | `unzip -o r2probe.zip` in the repo root → `outputs/teacher/r2probe/` | 32 GB |
-| `r2probe4k.zip` | same → `outputs/teacher/r2probe4k/` | 30 GB |
-
-Stored uncompressed (zip -0); paths are repo-relative.
-
-**What is inside each run folder.**
-- `bridge/<scene>/`: per scene (10 DL3DV training scenes, list `bridge/scenes.txt`), the memory side: `req_k` / `resp_k` = cameras and
-  our memory's raw renders at the clip's keyframes, `fix_k` = the same renders after the s200_paired fixer, `meta.json`.
-- `mem_g*/`: the memory runs' summaries (6 consecutive keyframes written: an early-episode memory).
-- `conditions/`: Gen3R's inputs per variant (conditioning frame + SDEdit start; variants realR, realF, RF, FF, RR).
-- `samples_s9/`, `samples_s7/` (r2probe only): Gen3R samples at SDEdit strength .9 / .7, 2 seeds, clips of 49 frames over the next
-  2 / 6 / 16 keyframes.
-- `probe2.json`, `probe2_regions.json`, `probe2_summary.json`: scores (PSNR, LPIPS, sharpness vs the real keyframes; covered / hole
-  regions); `run.sh` / `farm.sh`: the exact commands; `logs/`.
-- r2probe: frames from the 256 px orbit caches (`data/orbit_cache/dl3dv_train`); r2probe4k: the same 10 scenes from 4K video records
-  (`data/orbit_cache/dl3dv_video4k`).
-
-**Result in one line.** Gen3R never beat our fixed render (span 2, 4K: fixed render 14.12 dB / LPIPS .451 / 0.40x real sharpness; best
-Gen3R 14.06 / .477 / 0.26x; 5 % usable; span 6: 0 %), so Gen3R was dropped as a teacher on 2026-11-03.
-
-**To rerun or extend** (`scripts/lyra/teacher_probe2.py`, stages produce → memory → fix → convert → farm → score):
-- Gen3R code github.com/JaceyHuang/Gen3R at commit 1df25b6 in `third_party/gen3r`, weights huggingface.co/JaceyH919/Gen3R (MIT) in
-  `third_party/gen3r/checkpoints` (38 GB); its own conda env.
-- Our window model: `13_pixel_arm/model_001000.pt` + `upd_000800.pt` (the scripts load the model from
-  `checkpoints/flc_dev/12_pixel_arm_bk/model_001000.pt`, the identical file: place or link it there).
-- The fixer: `14_stream_fixer_s200_paired/`.
-- Source frames (not uploaded): `data/orbit_cache/dl3dv_train` (`scripts/data/build_orbit_cache.py`) and
-  `data/orbit_cache/dl3dv_video4k` (`scripts/data/build_video_cache.py`, from the DL3DV 4K training videos; ~3-4 h on 8 GPUs).
-
-**Terms.** Contains frames, renders and generated samples of DL3DV-10K training scenes (DL3DV terms of use apply) and Gen3R outputs
-(MIT). Released on the project owner's decision (2026-11-04).

16_gen3r_round2_probe/SHA256SUMS

@@ -1,3 +0,0 @@
-1e73c26b38fe5785f66b77a0386a1fb3a928a6e9a50ed8e8256c62a72ad86627  ./r2probe4k.zip
-bb0fdcf3a01e8e44ee207f30f0d7c86d17e7c2ce582f1de2ec8080de235f6f8d  ./r2probe.zip
-bdb1638859a33cb1883ef58052b9df9b3073a8bf320eded7845beecebcefdc7e  ./README.md

16_gen3r_round2_probe/r2probe.zip

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:bb0fdcf3a01e8e44ee207f30f0d7c86d17e7c2ce582f1de2ec8080de235f6f8d
-size 34338129643

16_gen3r_round2_probe/r2probe4k.zip

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:1e73c26b38fe5785f66b77a0386a1fb3a928a6e9a50ed8e8256c62a72ad86627
-size 31400347795

SHA256SUMS

@@ -1,8 +1,5 @@
-9a67a6036956c6d32cafe001e05e803f7c1f6895f5056052058cb58538eb8460  01_p4_mixed_v1.zip
-a68adbdd8bc8a097e91167d6361084d1cb743dedbafbf36d07fc38cf2a42b413  02_p4_control_v1.zip
 e67573c66ee8baeb892e6c32dc6a73380a68b4718d96d26e0bf9f0e738336aa9  03_checkpoints_streaming.zip
 e0056e2d7e12fe99eeb1161c0a633d7d7b18837ee12aa0dc3ca3bbf2f2a5db50  04_eval_data.zip
 14d454ea4e2caeb2f26478bf5661270b707c03b8f50640cdb2cb86029ba0797a  05_boards.zip
-fe7959a3bd5770cbd8075d34a97185c3fdf5aa1c7395f95b37403d1dcd622f42  06_gen3r_outputs.zip
 08e30bdaac8ddb560151384df06e16bc88606cbe86915bb92d587e16e522110e  13_pixel_arm/model_001000.pt
 7a1d6c3a76c68102f278048a8869ab589e4bab48e80250e9805eaaa428d99d5d  13_pixel_arm/upd_000800.pt