rename folders to public names

ControlGlaucoma_1208_base_v_prediction_noema/config.json

@@ -1,57 +0,0 @@
-{
-  "_class_name": "ControlNetModel",
-  "_diffusers_version": "0.31.0",
-  "_name_or_path": "/root/cloud/work_dirs/reward_model/10k/1128_v_prediction_8cja_fixed_timestep_segman/checkpoint-8000",
-  "act_fn": "silu",
-  "addition_embed_type": null,
-  "addition_embed_type_num_heads": 64,
-  "addition_time_embed_dim": null,
-  "attention_head_dim": [
-    5,
-    10,
-    20,
-    20
-  ],
-  "block_out_channels": [
-    320,
-    640,
-    1280,
-    1280
-  ],
-  "class_embed_type": null,
-  "conditioning_channels": 3,
-  "conditioning_embedding_out_channels": [
-    16,
-    32,
-    96,
-    256
-  ],
-  "controlnet_conditioning_channel_order": "rgb",
-  "cross_attention_dim": 1024,
-  "down_block_types": [
-    "CrossAttnDownBlock2D",
-    "CrossAttnDownBlock2D",
-    "CrossAttnDownBlock2D",
-    "DownBlock2D"
-  ],
-  "downsample_padding": 1,
-  "encoder_hid_dim": null,
-  "encoder_hid_dim_type": null,
-  "flip_sin_to_cos": true,
-  "freq_shift": 0,
-  "global_pool_conditions": false,
-  "in_channels": 4,
-  "layers_per_block": 2,
-  "mid_block_scale_factor": 1,
-  "mid_block_type": "UNetMidBlock2DCrossAttn",
-  "norm_eps": 1e-05,
-  "norm_num_groups": 32,
-  "num_attention_heads": null,
-  "num_class_embeds": null,
-  "only_cross_attention": false,
-  "projection_class_embeddings_input_dim": null,
-  "resnet_time_scale_shift": "default",
-  "transformer_layers_per_block": 1,
-  "upcast_attention": true,
-  "use_linear_projection": true
-}

ControlGlaucoma_1208_base_v_prediction_noema/diffusion_pytorch_model.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:69e69878599d7de163261f85b5da26fa5c532c3b3e2903b19cb2e690719dfb97
-size 1456953560

README.md

@@ -1,41 +0,0 @@
-# ControlGlaucoma Paper Checkpoint Backup
-
-All weights are ControlNetModel (StableDiffusion 2.1, v-prediction or epsilon).
-
-## Files
-
-### Generation used in paper (downstream classification)
-
-`ControlGlaucoma_v_prediction_seed42_ema/`
-  - Used to generate the 1296-image synth pool (paper Section 4.5 augmentation)
-  - Fine-tuned from `ControlGlaucoma_1208_base_v_prediction_ema` with CE-cup loss
-  - Trained 2026-04-18 with seed=42
-  - SD base: stabilityai/stable-diffusion-2-1 (v-prediction 768)
-  - Source: /root/cloud/work_dirs/reward_model/10k/4月新实验生成/
-           cecup_vp_seed42__0418_v_prediction_ulrj_reward_loss_cup_disc_loss1.1_fixed_timestep_segman_ema/
-
-### Intermediate base (for full lineage reproducibility)
-
-`ControlGlaucoma_1208_base_v_prediction_ema/`
-  - The initial ControlNet trained 2025-12-08, ID 5hf6
-  - Losses: reward + uncertainty + cup_disc_loss1.1 + fixed_timestep + SegMAN
-  - This is the starting point for the cecup_vp fine-tune above
-  - Source: /root/cloud/1208_v_prediction_5hf6_*_ema/
-
-### ε-prediction ablation variant
-
-`ControlGlaucoma_epsilon_seed42_ema/`
-  - ε-prediction version of CG, used for ablation in paper (Section 3 v vs ε)
-  - Same loss config as v-prediction
-  - Source: /root/cloud/work_dirs/reward_model/10k/4月新实验生成/
-           cecup_eps_mask_seed42__0421_epsilon_xf5e_reward_loss_cup_disc_loss1.1_fixed_timestep_segman_ema/
-
-## Generation workflow
-
-1. Load v-prediction CG ControlNet (cecup_vp_seed42_ema)
-2. Load SD 2.1 base (stock HuggingFace stabilityai/stable-diffusion-2-1)
-3. For each of 1296 healthy FairSeg images, generate multiple synth at different vCDR target scales
-4. Use SegMAN to estimate vCDR on each synth
-5. Split: vCDR > 0.6 → synth_glaucoma pool (2862 imgs)
-        vCDR < 0.6 → synth_health pool (7237 imgs)
-6. Use synth_glaucoma as positive-class augmentation in FairVLMed classification