Initial upload: ProgressiveDepth + RFTrans final ckpts and configs

README.md

@@ -1,3 +1,149 @@
-# GridDepth - placeholder
+---
+license: other
+language:
+- en
+tags:
+- depth-completion
+- transparent-objects
+- robotics
+- cleargrasp
+- lidf
+- rftrans
+- transdiff
+library_name: pytorch
+---
 
-Upload test from progressive depth bot.
+# GridDepth: Pretrained Checkpoints for Transparent-Object Depth Completion
+
+This repo hosts the **pretrained checkpoints** that go with the
+[atom525/ProgressiveDepth](https://github.com/atom525/ProgressiveDepth) codebase
+(idea.md series-joint pipeline: TransDiff Refined1 → LIDF) **plus** our local
+**RFTrans reproduction** baselines.
+
+> **Recipe**: see [atom525/ProgressiveDepth README.md](https://github.com/atom525/ProgressiveDepth)
+> and [docs/PIPELINE.md](https://github.com/atom525/ProgressiveDepth/blob/main/docs/PIPELINE.md).
+
+---
+
+## File layout
+
+```
+GridDepth/
+├── progressivedepth/                                # idea.md 主线（Module A=ip_basic + Module B=LIDF）
+│   ├── ckpts/
+│   │   ├── lidf_stage1_epoch059.pth                # 248 MB — LIDF Stage 1 (frozen baseline, CG-only Adam 60 ep)
+│   │   ├── C_stage2_epoch029.pth                   # 2.2 MB — Stage 2 RefineNet, retrained on Refined1 input (idea.md C run)
+│   │   └── C_stage3_epoch029.pth                   # 2.2 MB — Stage 3 RefineNet hard-neg, retrained on Refined1 input
+│   └── configs/
+│       ├── train_progressive_stage2.yaml
+│       ├── train_progressive_stage3.yaml
+│       └── pipeline_config.yaml                    # inference / evaluate config
+│
+└── rftrans/                                         # RFTrans 复现产物
+    ├── ckpts/
+    │   ├── rfnet_refractive_flow_epoch500.pth      # 467 MB — RFNet (DRN backbone), Adam 500 ep on unity/train
+    │   ├── f2net_flow2normal_epoch500.pth          # 356 MB — F2Net (simple_unet), Adam 500 ep on unity/train
+    │   ├── mask_adam_epoch195.pth                  # 312 MB — mask network (DRN), Adam 200 ep on unity/train, mIoU 0.847
+    │   └── outlines_side_adam_epoch195.pth         # 312 MB — boundary network (DRN side-output), Adam 200 ep on unity/train
+    └── configs/
+        ├── refractive_flow_config.yaml             # RFNet train config (Adam, 500 ep)
+        ├── flow2normal_config.yaml                 # F2Net train config (Adam, 500 ep)
+        ├── mask_adam_config.yaml                   # mask train config (Adam, 200 ep)
+        ├── outlines_side_adam_config.yaml          # boundary train config (Adam, 200 ep)
+        └── exp017_paperfaithful.yaml               # rgb2normal e2e config (paper-faithful: SGD 100 ep, lr=1e-4 mom=0.9 wd=5e-4)
+```
+
+---
+
+## ProgressiveDepth (idea.md series-joint pipeline)
+
+Pipeline:
+```
+RGB + Noisy Depth
+        │
+        ▼  Module A: TransDiff Data Preprocessing (ip_basic 多尺度形态学填充)
+   Refined Depth1
+        │
+        ▼  Module B: LIDF (Stage 1 frozen + Stage 2 / 3 retrained on Refined1)
+   Final Depth
+```
+
+### Final results (paper protocol: 256×144 + per-image avg + corrupt mask)
+
+C_full = `lidf_stage1_epoch059.pth` + `C_stage2_epoch029.pth` + `C_stage3_epoch029.pth`，evaluation 用 mode A (feed_to_lidf=refined1):
+
+| Dataset | C_full RMSE↓ | C_full δ1.05↑ | B baseline RMSE | B baseline δ1.05 | LIDF paper Table 1 |
+|---|---:|---:|---:|---:|---:|
+| **real-test (Real-novel)** ⭐ | **0.0403** | **45.28** | 0.0443 | 40.18 | 0.0250 / 76.21 |
+| real-val (Real-known) | 0.0351 | 77.22 | 0.0358 | 77.18 | 0.0280 / 82.37 |
+| synthetic-test (Syn-novel) | 0.0328 | 62.82 | 0.0305 | 66.12 | 0.0280 / 68.62 |
+| synthetic-val (Syn-known) | 0.0129 | 93.72 | 0.0111 | 96.07 | 0.0120 / 94.79 |
+
+**Conclusion**: idea.md series-joint approach is **effective on real-world data** (Real-novel RMSE ↓9%, δ1.05 ↑5 pts vs baseline B), **regression on synthetic** (where ip_basic adds noise to clean inputs). The remaining gap to paper Table 1 is due to Omniverse Object Dataset being unavailable (link broken since 2025-03, [NVlabs/implicit_depth#3](https://github.com/NVlabs/implicit_depth/issues/3)).
+
+---
+
+## RFTrans reproduction
+
+Pipeline (per RFTrans paper §III-C):
+```
+RGB ──> RFNet ──> refractive flow + mask + boundary
+                                          │
+                                          └──> F2Net ──> surface normal
+                                                              │
+                                                              └──> depth2depth global opt ──> Refined Depth
+```
+
+### Caveats
+
+1. **Architecture deviation**: paper §III-C says "RFNet predicts mask, boundary, and refractive flow" (multi-task), but the official repo doesn't implement this. We trained **separate networks** (RFNet predicts only flow, F2Net predicts normal from flow, mask & boundary as independent DeepLab+DRN networks) — this matches the actual repo structure but not the paper text.
+2. **Optimizer deviation**: paper §IV-A specifies SGD lr=1e-4 momentum=0.9 weight_decay=5e-4 for 100 epochs. We used **Adam** for sub-network training because we empirically found SGD lr=1e-4 from random init **does not converge** (mask val mIoU ~0.46 = random level after 100 ep SGD vs 0.85 with Adam 200 ep). The provided `exp017_paperfaithful.yaml` IS paper-faithful (SGD 100 ep) — used for the **end-to-end fine-tuning stage**, where it warm-starts from the Adam-trained RFNet/F2Net.
+3. **Training data**: all networks trained on `data/unity/train/` (5000 RGB + flow + mask + boundary + normal GT, generated with [Unity-RefractiveFlowRender](https://github.com/LJY-XCX/Unity-RefractiveFlowRender)) — this is the dataset specified by RFTrans paper §IV-A.
+
+### How to use these RFTrans ckpts
+
+In your `RFTrans/eval_depth_completion/config_*.yaml`:
+```yaml
+rgb2flow:
+  pathWeightsFile: <path_to>/rfnet_refractive_flow_epoch500.pth
+flow2normal:
+  pathWeightsFile: <path_to>/f2net_flow2normal_epoch500.pth
+masks:
+  pathWeightsFile: <path_to>/mask_adam_epoch195.pth        # OR cleargrasp_orig/.../checkpoint_mask.pth
+outlines:
+  pathWeightsFile: <path_to>/outlines_side_adam_epoch195.pth   # OR cleargrasp_orig/.../checkpoint_outlines.pth
+```
+
+---
+
+## Environment / dependencies
+
+- python 3.8, pytorch 2.0.0+cu118
+- LIDF: see [implicit_depth/requirements.txt](https://github.com/atom525/ProgressiveDepth/blob/main/implicit_depth/requirements.txt)
+- RFTrans: needs `depth2depth` C++ binary and `libhdf5.so` from conda env
+
+## License
+
+- LIDF Stage 1 ckpt and code: NVIDIA Source Code License (Non-Commercial), inherited from [NVlabs/implicit_depth](https://github.com/NVlabs/implicit_depth)
+- RFTrans ckpts and code: inherited from [LJY-XCX/RFTrans](https://github.com/LJY-XCX/RFTrans) license
+- Our extensions (transdiff_preprocess wrapper, train_progressive trainer, retrains): same as upstream
+
+## Citation
+
+If you use these ckpts please cite the original works:
+
+```bibtex
+@inproceedings{zhu2021rgbd,
+  title={RGB-D Local Implicit Function for Depth Completion of Transparent Objects},
+  author={Zhu, Luyang and Mousavian, Arsalan and Xiang, Yu and Mazhar, Hammad and van Eenbergen, Jozef and Debnath, Shoubhik and Fox, Dieter},
+  booktitle={CVPR},
+  year={2021}
+}
+
+@article{tang2024rftrans,
+  title={RFTrans: Leveraging Refractive Flow of Transparent Objects for Surface Normal Estimation and Manipulation},
+  author={Tang, Tutian and Liu, Jiyu and Zhang, Jieyi and Fu, Haoyuan and Xu, Wenqiang and Lu, Cewu},
+  journal={IEEE Robotics and Automation Letters},
+  year={2024}
+}
+```

progressivedepth/ckpts/C_stage2_epoch029.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7d54da2adc3492ac23902d19a586829e136bcc67d1d116ef741f4ba23ca7a554
+size 2272391

progressivedepth/ckpts/C_stage3_epoch029.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0d077188c8c85465169b1f346d7cb6bcf36ac7713389aff1f857c40d4141228e
+size 2272391

progressivedepth/ckpts/lidf_stage1_epoch059.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5db75336cfa1c8ea07251e37a93c17a7f9eeec11daa54615d41752e794446f15
+size 259974039

progressivedepth/configs/pipeline_config.yaml

@@ -0,0 +1,54 @@
+paths:
+  cleargrasp_root: /data1/liulingfeng/cooperation/ghy/ProgressiveDepth/implicit_depth/datasets/cleargrasp
+  lidf_stage1_ckpt: /data1/liulingfeng/cooperation/ghy/ProgressiveDepth/implicit_depth/logs/lidf/ckpt/bn24_lr0.001_nepo60_Adam_gres8_msn20000_vsn10000_rgb_resnet_embed_ROIAlign_pnet_twostage_offdec_IEF_niter2_probdec_IMNET_scatter_Maxpool_epo6_prob_ray_sn10.0_epo0_cleargrasp/epoch059_network.pth
+  lidf_stage2_ckpt: /data1/liulingfeng/cooperation/ghy/ProgressiveDepth/implicit_depth/logs/refine/ckpt/refine_bn24_lr0.001_nepo30_Adam_gres8_msn20000_vsn10000_refine_forward2_perturb0.8_s-0.2_e0.2_prob_ray_sn10.0_epo0_mixed/epoch029_network.pth
+  lidf_stage3_ckpt: /data1/liulingfeng/cooperation/ghy/ProgressiveDepth/implicit_depth/logs/refine/ckpt/refine_bn24_lr0.0001_nepo30_Adam_gres8_msn20000_vsn10000_refine_forward2_perturb0.8_s-0.2_e0.2_prob_ray_sn2.0_epo0_hardneg_cleargrasp_progressive_stage3_hardneg_refined1/epoch029_network.pth
+  output_dir: /data1/liulingfeng/cooperation/ghy/ProgressiveDepth/results
+transdiff:
+  max_depth: 100.0
+  blur_type: bilateral
+  extrapolate: false
+  feed_to_lidf: refined1   # idea.md 默认路径 A：TransDiff Refined1 喂给 LIDF
+lidf:
+  use_stage: stage3        # 论文 Table 1 数字对应的就是 LIDF + RefineNet (60 epoch + 30 hardneg)
+  miss_sample_num: 20000
+  hit_sample_num: 5000
+  forward_times: 2
+  device: cuda
+  gpu_id: 0
+evaluation:
+  datasets:
+  - real-test
+  - real-val
+  - synthetic-test
+  - synthetic-val
+  metrics:
+  - rmse
+  - abs_rel
+  - mae
+  - a1
+  - a2
+  - a3
+  min_depth: 0.0
+  max_depth: 100.0
+visualization:
+  save_vis: true
+  num_vis: 10
+  vis_types:
+  - noisy
+  - refined1
+  - final
+  - gt
+innovations:
+  use_mask_driven: false
+  mask_driven:
+    segment_loss_weight: 0.0
+    boundary_loss_weight: 0.0
+  use_spvnas: false
+  spvnas:
+    voxel_size_range:
+    - 4
+    - 16
+    sample_density_range:
+    - 8
+    - 32

progressivedepth/configs/train_progressive_stage2.yaml

@@ -0,0 +1,133 @@
+# ProgressiveDepth Stage 2 训练配置
+# - LIDF stage1 冻结，从已有 _cleargrasp 60-epoch ckpt 加载
+# - 数据加载阶段在 base dataset 外面套 TransDiffWrappedDataset，
+#   把 depth_corrupt 替换成 ip_basic Refined1，再喂给 LIDF
+# - 30 epoch + lr 0.001（与论文 stage2 完全对齐）
+# - 单 GPU（DDP off）以避免数据多副本对 Refined1 的额外 CPU 压力
+
+trainer_name: progressive
+exp_type: train
+base_log_dir: ../logs/refine
+lidf_ckpt_path: /data1/liulingfeng/cooperation/ghy/ProgressiveDepth/implicit_depth/logs/lidf/ckpt/bn24_lr0.001_nepo60_Adam_gres8_msn20000_vsn10000_rgb_resnet_embed_ROIAlign_pnet_twostage_offdec_IEF_niter2_probdec_IMNET_scatter_Maxpool_epo6_prob_ray_sn10.0_epo0_cleargrasp/epoch059_network.pth
+log_name:
+custom_postfix: 'progressive_stage2_refined1'
+resume: latest_network.pth
+gpu_id:
+vis_gpu: '5'
+
+mask_type: all
+
+# 我们的扩展：TransDiff Refined1 喂 LIDF 的开关
+progressive:
+  max_depth: 100.0
+  blur_type: bilateral
+  extrapolate: false
+  enable_train: true
+  enable_valid: true
+
+# 数据：因 Omniverse 缺失，只用 cleargrasp synthetic train
+dataset:
+  type: cleargrasp
+  cleargrasp_root_dir: ../datasets/cleargrasp
+  omniverse_root_dir: ../datasets/omniverse
+  use_data_augmentation: True
+  img_width: 320
+  img_height: 240
+  split_ratio: 0.9
+  omni_corrupt_all: True
+  corrupt_table: True
+  depth_aug: False
+  corrupt_all_pix: False
+  ellipse_dropout_mean: 20
+  ellipse_gamma_shape: 10.0
+  ellipse_gamma_scale: 1.0
+
+model:
+  rgb_model_type: resnet
+  rgb_embedding_type: ROIAlign
+  rgb_in: 3
+  rgb_out: 32
+  roi_inp_bbox: 8
+  roi_out_bbox: 2
+  pnet_model_type: twostage
+  pnet_in: 6
+  pnet_out: 128
+  pnet_gf: 32
+  pnet_pos_type: rel
+  pos_encode: True
+  intersect_pos_type: abs
+  multires: 8
+  multires_views: 4
+  offdec_type: IEF
+  n_iter: 2
+  probdec_type: IMNET
+  imnet_gf: 64
+  scatter_type: Maxpool
+  maxpool_label_epo: 0
+
+refine:
+  forward_times: 2
+  perturb: True
+  perturb_prob: 0.8
+  pnet_model_type: twostage
+  pnet_in: 6
+  pnet_out: 128
+  pnet_gf: 32
+  pnet_pos_type: rel
+  pos_encode: True
+  intersect_pos_type: abs
+  multires: 8
+  multires_views: 4
+  offdec_type: IEF
+  n_iter: 2
+  imnet_gf: 64
+  use_sigmoid: False
+  offset_range: [-0.2, 0.2]
+  use_all_pix: True
+
+grid:
+  res: 8
+  miss_sample_num: 20000
+  valid_sample_num: 10000
+  offset_range: [0., 1.]
+
+training:
+  batch_size: 24
+  valid_batch_size: 1
+  nepochs: 30
+  nepoch_decay: 30
+  decay_gamma: 0.1
+  nepoch_ckpt: 1
+  log_interval: 20
+  # 关掉可视化（同时被 train_progressive.visualize() override 兜底）
+  train_vis_iter: 999999
+  val_vis_iter: 999999
+  lr: 0.001
+  num_workers: 4
+  pin_memory: False
+  do_valid: True
+  valid_start_epo: 0
+  optimizer_name: Adam
+  scheduler_name: StepLR
+
+loss:
+  hard_neg: False
+  hard_neg_ratio:
+  pos_loss_type: single
+  pos_w: 100.0
+  prob_loss_type: ray
+  prob_w: 0
+  surf_norm_w: 10.0
+  surf_norm_epo: 0
+  smooth_w: 0
+  smooth_epo: 0
+
+dist:
+  ddp: False
+  dist_url:
+  dist_backend:
+  nodes_num: 1
+  node_rank: 0
+  ngpus_per_node: 1
+  world_size:
+  global_gpu_id:

progressivedepth/configs/train_progressive_stage3.yaml

@@ -0,0 +1,132 @@
+# ProgressiveDepth Stage 3 训练（hard-negative mining）—— 接续 stage2 retrain
+# 在 stage2 retrain 跑完（C run 给出 epoch_029 ckpt）后，用 ip_basic Refined1 输入
+# 续训 30 epoch hard-neg。
+#
+# 严格对齐论文 stage3 协议：lr=0.0001（lr/10），pos_w=20（/5），sn_w=2（/5），
+# loss.hard_neg=True, hard_neg_ratio=0.1（top 10% 误差像素）。
+# checkpoint_path 指向 stage2 retrain 的 epoch_029，避免从 noisy-trained ckpt 暖启。
+
+trainer_name: progressive
+exp_type: train
+base_log_dir: ../logs/refine
+lidf_ckpt_path: /data1/liulingfeng/cooperation/ghy/ProgressiveDepth/implicit_depth/logs/lidf/ckpt/bn24_lr0.001_nepo60_Adam_gres8_msn20000_vsn10000_rgb_resnet_embed_ROIAlign_pnet_twostage_offdec_IEF_niter2_probdec_IMNET_scatter_Maxpool_epo6_prob_ray_sn10.0_epo0_cleargrasp/epoch059_network.pth
+checkpoint_path: /data1/liulingfeng/cooperation/ghy/ProgressiveDepth/implicit_depth/logs/refine/ckpt/refine_bn24_lr0.001_nepo30_Adam_gres8_msn20000_vsn10000_refine_forward2_perturb0.8_s-0.2_e0.2_prob_ray_sn10.0_epo0_cleargrasp_progressive_stage2_refined1/epoch029_network.pth
+log_name:
+custom_postfix: 'progressive_stage3_hardneg_refined1'
+resume: latest_network.pth
+gpu_id:
+vis_gpu: '5'
+
+mask_type: all
+
+progressive:
+  max_depth: 100.0
+  blur_type: bilateral
+  extrapolate: false
+  enable_train: true
+  enable_valid: true
+
+dataset:
+  type: cleargrasp
+  cleargrasp_root_dir: ../datasets/cleargrasp
+  omniverse_root_dir: ../datasets/omniverse
+  use_data_augmentation: True
+  img_width: 320
+  img_height: 240
+  split_ratio: 0.9
+  omni_corrupt_all: True
+  corrupt_table: True
+  depth_aug: False
+  corrupt_all_pix: False
+  ellipse_dropout_mean: 20
+  ellipse_gamma_shape: 10.0
+  ellipse_gamma_scale: 1.0
+
+model:
+  rgb_model_type: resnet
+  rgb_embedding_type: ROIAlign
+  rgb_in: 3
+  rgb_out: 32
+  roi_inp_bbox: 8
+  roi_out_bbox: 2
+  pnet_model_type: twostage
+  pnet_in: 6
+  pnet_out: 128
+  pnet_gf: 32
+  pnet_pos_type: rel
+  pos_encode: True
+  intersect_pos_type: abs
+  multires: 8
+  multires_views: 4
+  offdec_type: IEF
+  n_iter: 2
+  probdec_type: IMNET
+  imnet_gf: 64
+  scatter_type: Maxpool
+  maxpool_label_epo: 0
+
+refine:
+  forward_times: 2
+  perturb: True
+  perturb_prob: 0.8
+  pnet_model_type: twostage
+  pnet_in: 6
+  pnet_out: 128
+  pnet_gf: 32
+  pnet_pos_type: rel
+  pos_encode: True
+  intersect_pos_type: abs
+  multires: 8
+  multires_views: 4
+  offdec_type: IEF
+  n_iter: 2
+  imnet_gf: 64
+  use_sigmoid: False
+  offset_range: [-0.2, 0.2]
+  use_all_pix: True
+
+grid:
+  res: 8
+  miss_sample_num: 20000
+  valid_sample_num: 10000
+  offset_range: [0., 1.]
+
+training:
+  batch_size: 24
+  valid_batch_size: 1
+  nepochs: 30
+  nepoch_decay: 30
+  decay_gamma: 0.1
+  nepoch_ckpt: 1
+  log_interval: 20
+  train_vis_iter: 999999
+  val_vis_iter: 999999
+  lr: 0.0001
+  num_workers: 4
+  pin_memory: False
+  do_valid: True
+  valid_start_epo: 0
+  optimizer_name: Adam
+  scheduler_name: StepLR
+
+loss:
+  hard_neg: True
+  hard_neg_ratio: 0.1
+  pos_loss_type: single
+  pos_w: 20.0
+  prob_loss_type: ray
+  prob_w: 0
+  surf_norm_w: 2.0
+  surf_norm_epo: 0
+  smooth_w: 0
+  smooth_epo: 0
+
+dist:
+  ddp: False
+  dist_url:
+  dist_backend:
+  nodes_num: 1
+  node_rank: 0
+  ngpus_per_node: 1
+  world_size:
+  global_gpu_id:

rftrans/ckpts/f2net_flow2normal_epoch500.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:96b61fe6255bb985d39513a7c24deb353519d66cf3a1b81877f03754dd14365d
+size 372605797

rftrans/ckpts/mask_adam_epoch195.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1a63181d9ca1a1e4f48a5a9bc6cb563367246b3716df54f56cda401ae948ba5b
+size 326305765

rftrans/ckpts/outlines_side_adam_epoch195.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0781611ccaa86c04012ba366ff8e8cf51039fee4eeb330527d3bc5bcb4a7db27
+size 326308517

rftrans/ckpts/rfnet_refractive_flow_epoch500.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a5b75cf2233aab6a637c4a5676ebb3562489bea1e35423c703fd72a8e34dbc03
+size 489345509

rftrans/configs/exp017_paperfaithful.yaml

@@ -0,0 +1,109 @@
+# RFTrans Table II reproduction - end-to-end fine-tune with paper-faithful loss
+# Bug fix: training previously used loss = L_flow + alpha*L_norm with alpha=0.02
+# (flow dominates ~100x, normal supervision essentially ignored).
+# Paper Eq.: L = alpha * L_flow + L_norm with alpha=0.01 (best per Table IV).
+train:
+  datasetsTrain:
+    - images: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/train_cg/RGB'
+      flows: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/train_cg/flow'
+      labels: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/train_cg/normal'
+      masks: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/train_cg/mask'
+
+  datasetsVal:
+    - images: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/valid/RGB'
+      flows: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/valid/flow'
+      labels: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/valid/normal'
+      masks: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/valid/mask'
+
+  datasetsTestReal:
+  datasetsTestSynthetic:
+
+  rgb2flow:
+    model: "drn"
+    numClasses: 2
+    numInputChannels: 3
+    pathWeightsFile: "/data1/liulingfeng/cooperation/ghy/RFTrans-repro/RFTrans/pytorch_networks/refractive_flow/logs-deeplab/exp-001/checkpoints/checkpoint-epoch-0500.pth"
+
+  flow2normal:
+    model: "simple_unet"
+    numClasses: 3
+    numInputChannels: 3
+    pathWeightsFile: "/data1/liulingfeng/cooperation/ghy/RFTrans-repro/RFTrans/pytorch_networks/flow2normal/logs-deeplab/exp-000/checkpoints/checkpoint-epoch-0500.pth"
+
+  model: "drn"
+  batchSize: 8
+  validationBatchSize: 8
+  testBatchSize: 8
+  numEpochs: 100
+  # Train at data's native 512x512 (matches the pretrained R2F/F2N which learned
+  # flow magnitudes in the 512-frame). Paper says 256x256 but our train_cg data
+  # is rendered at 512x512 and the pretrained checkpoints were trained at 512.
+  imgHeight: 512
+  imgWidth: 512
+  numWorkers: 8
+  logsDir: "logs-deeplab"
+  lossFunc: "cosine"
+  percentageDataForTraining: 1.0
+  percentageDataForValidation: 0.5
+
+  outputStride: 8
+  epochSize: 1
+
+  # initialize from previously trained separate checkpoints (RFNet / F2Net)
+  continueTraining: True
+  pathPrevCheckpoint: ""
+  initOptimizerFromCheckpoint: False
+  loadEpochNumberFromCheckpoint: False
+
+  saveImageInterval: 1
+  saveImageIntervalIter: 1000
+  testInterval: 1
+  saveModelInterval: 1
+
+  # NOTE: paper specifies SGD lr=1e-4 momentum 0.9 weight_decay 5e-4 for 100 ep,
+  # but fine-tuning from already-Adam-trained pretrained R2F/F2N with SGD@1e-4
+  # barely moves the weights. Use Adam @ 1e-4 to converge faster from the
+  # pretrained init (loss bug previously suppressed the normal supervision).
+  optimizer: "SGD"
+  optimSgd:
+    learningRate: 1e-4
+    momentum: 0.9
+    weight_decay: 5e-4
+  optimAdam:
+    learningRate: 1e-4
+    weightDecay: 0.0001
+  lrScheduler: "StepLR"
+  lrSchedulerStep:
+    step_size: 30
+    gamma: 0.5
+
+  # alpha for joint loss; paper best = 0.01 with L = alpha * L_flow + L_norm
+  loss_alpha: 0.01
+
+eval:
+  datasetsSynthetic:
+    - images: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/valid/RGB'
+      flows: ''
+      labels: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/valid/normal'
+      masks: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/valid/mask'
+
+  datasetsReal:
+
+  rgb2flow:
+    model: "drn"
+    numClasses: 2
+    numInputChannels: 3
+    pathWeightsFile: ""
+
+  flow2normal:
+    model: "simple_unet"
+    numClasses: 3
+    numInputChannels: 3
+    pathWeightsFile: ""
+
+  model: "drn"
+  batchSize: 4
+  imgHeight: 256
+  imgWidth: 256
+  numWorkers: 4
+  resultsDir: "data/results"

rftrans/configs/flow2normal_config.yaml

@@ -0,0 +1,120 @@
+# train.py Config - Training
+train:
+  # For datasets, please pass atleast 1 value. If no datasets exist, pass "" as path for images.
+  # Synthetic datasets with ground truth labels
+  datasetsTrain:
+
+  - images: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/train/flow'
+    labels: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/train/normal'
+    masks: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/train/mask'
+
+#  - images: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/train/predicted_flows'
+#    labels: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/train/normal'
+#    masks: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/train/mask'
+
+
+  # Synthetic datasets with ground truth labels - 10% split of train
+  datasetsVal:
+    - images: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/valid/flow'
+      labels: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/valid/normal'
+      masks: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/valid/mask'
+#
+#    - images: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/valid/predicted_flows'
+#      labels: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/valid/normal'
+#      masks: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/valid/mask'
+
+  # Real Images (no ground truth labels)
+  datasetsTestReal:
+    # - images: 'data/datasets/test/camera-pics/resized-files/preprocessed-rgb-imgs'
+    #   labels: ''
+    # - images: 'data/datasets/test/realsense-captures/resized-files/preprocessed-rgb-imgs'
+    #   labels: ''
+    # - images: 'data/datasets/test/realsense-demo-table-2/source-files/rgb-imgs'
+    #   labels: ''
+    # - images: 'data/datasets/test/realsense-demo-table-3/source-files/rgb-imgs'
+    #   labels: ''
+
+  # Synthetic datasets with ground truth labels - Used as test set
+  datasetsTestSynthetic:
+#    - images: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/valid/calibration/refractive_flow'
+#      labels: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/valid/normal'
+
+  # Training/Validation Params
+  model: "simple_unet" # Possible values: ["drn", unet", "deeplab_xception", "deeplab_resnet", "refinenet", "simple_unet"]
+  batchSize: 16
+  batchSizeMatterport: 0
+  batchSizeScannet: 0
+  validationBatchSize: 16
+  testBatchSize: 16
+  numEpochs: 501
+  imgHeight: 512
+  imgWidth: 512
+  numClasses: 3
+  numInputChannels: 3 # Num of channels in input image. RGB = 3 channels, Grayscale = 1 channel.
+  numWorkers: 16 # Num of workers used in the dataloader
+  logsDir: "logs-deeplab" # Directory where logs of each exp will be saved.
+  lossFunc: "cosine" # Possible values: ['cosine', 'radians']
+  percentageDataForTraining: 1.0 # The percentage of images in dataset to be used for training.
+  percentageDataForMatterportTraining: 0.5 # The percentage of images in dataset to be used for training.
+  percentageDataForScannetTraining: 0.35
+  percentageDataForValidation: 0.25
+  percentageDataForMatterportVal: 0.5
+  percentageDataForScannettVal: 0.5
+
+  # Deeplab specific
+  outputStride: 8 # Possible values: [8, 16]. Output stride for deeplabv3 model. Smaller values give finer details in output mask.
+  epochSize: 1 # After these many epochs, change learning rate
+
+  continueTraining: False # If true, continue training from a checkpoint
+  pathPrevCheckpoint: "/home/jiyu/ClearGrasp/pytorch_networks/refraction_flows/logs-deeplab/exp-003/checkpoints-adam/checkpoint-epoch-0020.pth" # Path to .pth checkpoint file to load to continue training from
+  initOptimizerFromCheckpoint: False # Re-Initialize optimizer's state from checkpoint. NOTE: when this is enabled, value of learningRate will be overridden with value from checkpoint.
+  loadEpochNumberFromCheckpoint: False # If true, the epoch/iter numbering will start from the checkpoint's last epoch num.
+
+  saveImageInterval: 1 # Log output images to tensorboard every saveImageInterval epochs
+  saveImageIntervalIter: 100 # Every N iterations, log output images to tensorboard
+  testInterval: 1 # Run on test set every nTestInterval epochs. Keep at 0 to skip tests.
+  saveModelInterval: 10 # Save the model checkpoints every N epochs
+
+  # Optimizer Params
+  optimizer: "Adam" # Possible Values: ["SGD", "Adam"]
+  optimAdam:
+    learningRate: 5e-5
+    weightDecay: 0.0001 # Other values: 0.0001
+  optimSgd:
+    learningRate: 1e-6
+    momentum: 0.9
+    weight_decay: 5e-4
+  lrScheduler: "StepLR" # Possible Values: ['', 'StepLR', 'ReduceLROnPlateau']
+  lrSchedulerStep:
+    step_size: 1000
+    gamma: 0.1
+  lrSchedulerPlateau:
+    factor: 0.8
+    patience: 25
+    verbose: True
+
+# eval.py Config - Validation/Testing Inference
+eval:
+  # Synthetic datasets with ground truth labels
+  # Used as validation set
+  datasetsSynthetic:
+    - images: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/valid/flow'
+      labels: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/valid/normal'
+      masks: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/valid/mask'
+
+  # Datasets of real images, no labels available
+  # Used as Test set
+  datasetsReal:
+
+
+  # Params
+  model: "simple_unet" # Possible values: ["drn", unet", "deeplab_xception", "deeplab_resnet", "refinenet", "simple_unet"]
+  numClasses: 3
+  batchSize: 16
+  imgHeight: 512
+  imgWidth: 512
+  os: 8
+  numWorkers: 4 # Num of workers used in the dataloader
+  pathWeightsFile: "/home/jiyu/ClearGrasp/pytorch_networks/flow2normal/logs-deeplab/exp-012/checkpoints/checkpoint-epoch-0500.pth" # Path to the checkpoint to be loaded
+  resultsDir: "data/results/"
+

rftrans/configs/mask_adam_config.yaml

@@ -0,0 +1,76 @@
+# RFTrans mask network training (binary semantic segmentation: transparent vs background)
+# 数据：unity/train/RGB + unity/train/mask（5000 张，512×512）
+# 架构：DeepLab + DRN backbone + cross_entropy2d，跟 occlusion_boundaries 一致
+# 目标：替换掉 ClearGrasp 2019 老 ckpt，看是否能改进 RFTrans pipeline 的最终 RMSE
+train:
+  # 训练集
+  datasetsTrain:
+    - images: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/train/RGB'
+      labels: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/train/mask'
+
+  # 验证集（10% 切出）
+  datasetsVal:
+    - images: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/valid/RGB'
+      labels: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/valid/mask'
+
+  # 不用 matterport / scannet / 真实集
+  datasetsMatterportTrain:
+  datasetsMatterportVal:
+  datasetsScannetTrain:
+  datasetsScannetVal:
+  datasetsTestReal:
+  datasetsTestSynthetic:
+    - images: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/valid/RGB'
+      labels: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/valid/mask'
+
+  # 训练超参（与 occlusion_boundaries 默认对齐）
+  model: "drn"
+  batchSize: 32
+  batchSizeMatterport: 0
+  batchSizeScannet: 0
+  validationBatchSize: 8
+  testBatchSize: 8
+  numEpochs: 200             # mask 任务相对简单，200 epoch 够
+  imgHeight: 256
+  imgWidth: 256
+  numClasses: 2              # binary: bg(0) / transparent(1)
+  numInputChannels: 3
+  numWorkers: 8
+  logsDir: "logs-deeplab"
+  lossFunc: "ce"             # cross-entropy 2d，模型用 deeplab/drn 时 train.py 走 utils.cross_entropy2d
+  percentageDataForTraining: 1.0
+  percentageDataForMatterportTraining: 0.5
+  percentageDataForScannetTraining: 0.35
+  percentageDataForValidation: 1.0
+  percentageDataForMatterportVal: 0.5
+  percentageDataForScannettVal: 0.5
+
+  outputStride: 8
+  epochSize: 1
+
+  continueTraining: False
+  pathPrevCheckpoint: ""
+  initOptimizerFromCheckpoint: False
+  loadEpochNumberFromCheckpoint: False
+
+  saveImageInterval: 5
+  saveImageIntervalIter: 200
+  testInterval: 1
+  saveModelInterval: 5
+
+  optimizer: "Adam"
+  optimAdam:
+    learningRate: 0.0001
+    weightDecay: 0
+  optimSgd:
+    learningRate: 1.0e-6
+    momentum: 0.9
+    weight_decay: 5.0e-4
+  lrScheduler: "StepLR"
+  lrSchedulerStep:
+    step_size: 20
+    gamma: 0.95
+  lrSchedulerPlateau:
+    factor: 0.8
+    patience: 25
+    verbose: True

rftrans/configs/outlines_side_adam_config.yaml

@@ -0,0 +1,130 @@
+# train.py Config - Training
+train:
+  # For datasets, please pass atleast 1 value. If no datasets exist, pass "" as path for images.
+  # Synthetic datasets with ground truth labels
+  datasetsTrain:
+  - images: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/train/RGB'
+    labels: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/train/boundary'
+
+
+  # Synthetic datasets with ground truth labels - 10% split of train
+  datasetsVal:
+    - images: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/valid/RGB'
+      labels: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/valid/boundary'
+
+
+  datasetsMatterportTrain:
+    # - images: 'data/datasets/matterport3d/train/matterport_rgb/v1/scans'
+    #   labels: 'data/datasets/matterport3d/train/matterport_render_normal'
+  datasetsMatterportVal:
+    # - images: 'data/datasets/matterport3d/val/matterport_rgb/v1/scans'
+    #   labels: 'data/datasets/matterport3d/val/matterport_render_normal'
+  datasetsScannetTrain:
+    # - images: 'data/datasets/scannet/scannet-rgb/scans/train'
+    #   labels: 'data/datasets/scannet/scannet_render_normal/train'
+  datasetsScannetVal:
+    # - images: 'data/datasets/scannet/scannet-rgb/scans/val'
+    #   labels: 'data/datasets/scannet/scannet_render_normal/val'
+
+  # Real Images (no ground truth labels)
+  datasetsTestReal:
+    # - images: 'data/datasets/test/camera-pics/resized-files/preprocessed-rgb-imgs'
+    #   labels: ''
+    # - images: 'data/datasets/test/realsense-captures/resized-files/preprocessed-rgb-imgs'
+    #   labels: ''
+    # - images: 'data/datasets/test/realsense-demo-table-2/source-files/rgb-imgs'
+    #   labels: ''
+    # - images: 'data/datasets/test/realsense-demo-table-3/source-files/rgb-imgs'
+    #   labels: ''
+
+  # Synthetic datasets with ground truth labels - Used as test set
+  datasetsTestSynthetic:
+    - images: 'data/data/datasets/test-synthetic/scoop-val/source-files/rgb-imgs'
+      labels: 'data/data/datasets/test-synthetic/scoop-val/source-files/outlines'
+
+  # Training/Validation Params
+  model: "drn" # Possible values: ['deeplab_xception', 'deeplab_resnet', 'drn']
+  batchSize: 64
+  batchSizeMatterport: 0
+  batchSizeScannet: 0
+  validationBatchSize: 8
+  testBatchSize: 8
+  numEpochs: 200
+  imgHeight: 256
+  imgWidth: 256
+  numClasses: 3
+  numInputChannels: 3 # Num of channels in input image. RGB = 3 channels, Grayscale = 1 channel.
+  numWorkers: 8 # Num of workers used in the dataloader
+  logsDir: "logs-deeplab-side" # Directory where logs of each exp will be saved.
+  lossFunc: "cosine" # Possible values: ['cosine', 'radians']
+  percentageDataForTraining: 1.0 # The percentage of images in dataset to be used for training.
+  percentageDataForMatterportTraining: 0.5 # The percentage of images in dataset to be used for training.
+  percentageDataForScannetTraining: 0.35
+  percentageDataForValidation: 1.0
+  percentageDataForMatterportVal: 0.5
+  percentageDataForScannettVal: 0.5
+
+  # Deeplab specific
+  outputStride: 8 # Possible values: [8, 16]. Output stride for deeplabv3 model. Smaller values give finer details in output mask.
+  epochSize: 1 # After these many epochs, change learning rate
+
+  continueTraining: True # If true, continue training from a checkpoint
+  pathPrevCheckpoint: "/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/cleargrasp_orig/cleargrasp-checkpoints/outlines/checkpoint_outlines.pth" # Path to .pth checkpoint file to load to continue training from
+  initOptimizerFromCheckpoint: False # Re-Initialize optimizer's state from checkpoint. NOTE: when this is enabled, value of learningRate will be overridden with value from checkpoint.
+  loadEpochNumberFromCheckpoint: False # If true, the epoch/iter numbering will start from the checkpoint's last epoch num.
+
+  saveImageInterval: 1 # Log output images to tensorboard every saveImageInterval epochs
+  saveImageIntervalIter: 100 # Every N iterations, log output images to tensorboard
+  testInterval: 1 # Run on test set every nTestInterval epochs. Keep at 0 to skip tests.
+  saveModelInterval: 5 # Save the model checkpoints every N epochs
+
+  # Optimizer Params
+  optimAdam:
+    learningRate: 0.0001
+    weightDecay: 0 # Other values: 0.0001
+  optimSgd:
+    learningRate: 1e-7
+    momentum: 0.9
+    weight_decay: 5e-4
+  lrScheduler: "StepLR" # Possible Values: ['', 'StepLR', 'ReduceLROnPlateau']
+  lrSchedulerStep:
+    step_size: 7
+    gamma: 0.1
+  lrSchedulerPlateau:
+    factor: 0.8
+    patience: 25
+    verbose: True
+
+# eval.py Config - Validation/Testing Inference
+eval:
+  # Synthetic datasets with ground truth labels
+  # Used as validation set
+  datasetsSynthetic:
+    - images: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/train_rgmask/RGB'
+      labels: ''
+
+  # Datasets of real images, no labels available
+  # Used as Test set
+  datasetsReal:
+    # - images: "datasets-transparent/studio_pics_sorted/selected_test/d415"
+    #   labels: "datasets-transparent/studio_pics_sorted/selected_test/d415"
+    # - images: "datasets-transparent/studio_pics_sorted/selected_test/d435"
+    #   labels: "datasets-transparent/studio_pics_sorted/selected_test/d435"
+    # - images: "datasets-transparent/studio_pics_sorted/selected_val/d435"
+    #   labels: "datasets-transparent/studio_pics_sorted/selected_val/d435"
+
+  datasetsMatterport:
+    # - images: 'data/datasets/matterport3d/train/matterport_rgb/v1/scans'
+    #   labels: 'data/datasets/matterport3d/train/matterport_render_normal'
+
+  # Params
+  model: "drn" # Possible values: ['deeplab_xception', 'deeplab_resnet', 'drn']
+  numClasses: 3
+  batchSize: 32
+  imgHeight: 256
+  imgWidth: 256
+  os: 8
+  numWorkers: 4 # Num of workers used in the dataloader
+  pathWeightsFile: "/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/cleargrasp_orig/cleargrasp-checkpoints/outlines/checkpoint_outlines.pth" # Path to the checkpoint to be loaded
+  resultsDir: "data/results"
+

rftrans/configs/refractive_flow_config.yaml

@@ -0,0 +1,100 @@
+# train.py Config - Training
+train:
+  # For datasets, please pass atleast 1 value. If no datasets exist, pass "" as path for images.
+  # Synthetic datasets with ground truth labels
+  datasetsTrain:
+  - images: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/train/RGB'
+    labels: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/train/flow'
+    masks: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/train/mask'
+
+  # Synthetic datasets with ground truth labels - 10% split of train
+  datasetsVal:
+    - images: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/valid/RGB'
+      labels: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/valid/flow'
+      masks: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/valid/mask'
+
+  # Real Images (no ground truth labels)
+  datasetsTestReal:
+
+  # Synthetic datasets with ground truth labels - Used as test set
+  datasetsTestSynthetic:
+
+  # Training/Validation Params
+  model: "drn" # Possible values: ['deeplab_xception', 'deeplab_resnet', 'drn']
+  batchSize: 24
+  batchSizeMatterport: 0
+  batchSizeScannet: 0
+  validationBatchSize: 64
+  testBatchSize: 16
+  numEpochs: 501
+  imgHeight: 512
+  imgWidth: 512
+  numClasses: 2
+  numInputChannels: 3 # Num of channels in input image. RGB = 3 channels, Grayscale = 1 channel.
+  numWorkers: 4 # Num of workers used in the dataloader
+  logsDir: "logs-deeplab" # Directory where logs of each exp will be saved.
+  #lossFunc: "cosine" # Possible values: ['cosine', 'radians']
+  percentageDataForTraining: 1.0 # The percentage of images in dataset to be used for training.
+  percentageDataForMatterportTraining: 0.5 # The percentage of images in dataset to be used for training.
+  percentageDataForScannetTraining: 0.35
+  percentageDataForValidation: 1.0
+  percentageDataForMatterportVal: 0.5
+  percentageDataForScannettVal: 0.5
+
+  # Deeplab specific
+  outputStride: 8 # Possible values: [8, 16]. Output stride for deeplabv3 model. Smaller values give finer details in output mask.
+  epochSize: 1 # After these many epochs, change learning rate
+
+  continueTraining: False # If true, continue training from a checkpoint
+  pathPrevCheckpoint: "" # Path to .pth checkpoint file to load to continue training from
+  initOptimizerFromCheckpoint: False # Re-Initialize optimizer's state from checkpoint. NOTE: when this is enabled, value of learningRate will be overridden with value from checkpoint.
+  loadEpochNumberFromCheckpoint: True # If true, the epoch/iter numbering will start from the checkpoint's last epoch num.
+
+  saveImageInterval: 1 # Log output images to tensorboard every saveImageInterval epochs
+  saveImageIntervalIter: 1000 # Every N iterations, log output images to tensorboard
+  testInterval: 10 # Run on test set every nTestInterval epochs. Keep at 0 to skip tests.
+  saveModelInterval: 10 # Save the model checkpoints every N epochs
+
+  # Optimizer Params
+  optimizer: "Adam" # Possible Values: ["SGD", "Adam"]
+  optimAdam:
+    learningRate: 5e-5  # Origin value: 0.0001
+    weightDecay: 0 # Other values: 0.0001
+  optimSgd:
+    learningRate: 1e-6
+    momentum: 0.9
+    weight_decay: 5e-4
+  lrScheduler: "StepLR" # Possible Values: ['', 'StepLR', 'ReduceLROnPlateau'] 
+  lrSchedulerStep:
+    step_size: 1000
+    gamma: 0.1
+  lrSchedulerPlateau:
+    factor: 0.8
+    patience: 25
+    verbose: True
+
+# eval.py Config - Validation/Testing Inference
+eval:
+  # Synthetic datasets with ground truth labels
+  # Used as validation set
+  datasetsSynthetic:
+    - images: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/valid/RGB'
+      labels: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/valid/flow'
+      masks: '/data1/liulingfeng/cooperation/ghy/RFTrans-repro/data/unity/valid/mask'
+
+  # Datasets of real images, no labels available
+  # Used as Test set
+  datasetsReal:
+
+
+  # Params
+  model: "drn" # Possible values: ['deeplab_xception', 'deeplab_resnet', 'drn']
+  numClasses: 2
+  batchSize: 64
+  imgHeight: 512
+  imgWidth: 512
+  os: 8
+  numWorkers: 4 # Num of workers used in the dataloader
+  pathWeightsFile: "/home/jiyu/ClearGrasp/pytorch_networks/refractive_flow/logs-deeplab/exp-033/checkpoints/checkpoint-epoch-0500.pth" # Path to the checkpoint to be loaded
+  resultsDir: "data/results"
+