Remove sample inputs, already cloned locally by user

Push FORGE3DBench FB150 sample inputs (150 objects x 4 views) for later testing

Add files using upload-large-folder tool

Add Amodal3R align_v2 mesh outputs (toys4k300 1v/4v, omni100 4v, FB150 maskfix+original, rest658 maskfix) + cameras

Remove stray .pyc added by previous commit

Amodal3R wrapper fixes: occlusion-aware FORGE3DBench mask in batch_amodal3r.py

Remove log / cache / raw-output files flagged by the 2026-09-30 public-repo audit (category a)

Remove slat_ckpt_sweep result json files (data safely migrated to private forge3d_release)

Remove results/metrics content (migrated to private forge3d_release/bert_simpson_migrated); code and checkpoints untouched

Add Ours meshes: aws_20260926/meshes/ours/toys4k300_ablation (tars + README + MANIFEST)

Add Ours meshes: aws_20260926/meshes/ours/toys4k100_slat_sweep (tars + README + MANIFEST)

Add Ours meshes: aws_20260926/meshes/ours/fb150_rest658 (tars + README + MANIFEST)

AWS 4-view benchmark baseline meshes: pixal3d_official50k (raw + placed .tar)

AWS 4-view benchmark baseline meshes: pixal3d_recenter_origin_4v (raw + placed .tar)

AWS 4-view benchmark baseline meshes: reconviagen_4v (raw + placed .tar)

AWS 4-view benchmark baseline meshes: hy3d2mv_4v (raw + placed .tar)

AWS 4-view benchmark baseline meshes: cupid_4v (raw + placed .tar)

AWS 4-view benchmark baseline meshes: amodal3r_4v (raw + placed .tar)

aws_20260926: Pixal3D official-code 50k geometry scores (FB150, rest658) + changed Ours batched cell script

aws_20260926: slatflow_mv8_200k_from32k final state step 62500 (model+EMA) + logs + README

Remove session results (moved to private forge3d_release)

session results 2026-09-26..29 (pixal3d reruns, texture bake, depth-source ablation, naive sam3d 1v, decimate30k, visuals, FINAL tables)

AWS run eval outputs: Toys4K-300 4v ablation (native + 512), Toys4K-100 SLAT ckpt sweep, FB150 per-object

SLAT flow step_0055000 (EMA)

SLAT flow step_0055000 (model)

SLAT flow step_0045000 (EMA)

SLAT flow step_0045000 (model)

SLAT flow step_0035000 (EMA)

SLAT flow step_0035000 (model)

AWS run code: run_toys4k300.sh summarize fix + abl ssonly; SS-only ablation row; 512 eval wrappers; official Pixal3D MV runner (GT poses); results tools

Pixal3D official MV (50k faces) FORGE3DBench 4v: RAW meshes (+ inputs transforms.json), status, logs

FORGE3DBench rest658 4v: per-object results (CSV + raw eval rows) for all 6 methods, asset-level means

FORGE3DBench 4v results: add rest658 held-out NV + weighted (4:24) for all methods

FORGE3DBench rest658: held-out views (all non-input cameras with visibility), index, wrapper, README addendum

FORGE3DBench 4v results (AWS run): IV/NV/weighted LPIPS-PSNR-SSIM + CD-L2/CD-L1/F-score, ours + 5 baselines

Add Toys4K-300 evaluation set (inputs tar, guide, view/ablation recipes, code)

FORGE3DBench: final eval protocol + eval_final.py, batched Ours inference, held-out view tars, missing-object lists, README

Upload ckpt/ssflow_seedcond_v3_20260916/step_0050000_ema.pt with huggingface_hub

Add SS-flow seedcond v3 step 50k checkpoint (raw weights)

Add Toys4K-100 4-view evaluation set (tar + docs + eval code)

forgebench: final Omni VGGT-seed ablation scripts (v2 registration, v3 mask filter, ours_incr_var driver, scoring, table rows)

SLAT flow step_0060000 (model+EMA)

forgebench: final scorer tree (shard support), aggregate_results.py, latest Omni VGGT drivers (v2/v3), README updates; no seeds

SLAT flow step_0050000 (model+EMA)

SLAT flow step_0040000 (model+EMA)

forgebench: Omni3D-300 GT canonical meshes as a single tar

forgebench: cropK Pixal3D fix, cammap_fb150, scoring wrappers, Omni real-image VGGT code (creds removed), FB150 + Omni 4v input tars, README

forgebench README: SS-flow 80k Hub path

Add files using upload-large-folder tool

forgebench README: Cupid align_v2, 4v inference section, Pixal3D 4v hold warning