FileNotFoundError during testing (missing Real/test *_label.pkl) and mismatch between segmentation result keys and code

[yumomo11]

Summary

I’m a sophomore undergraduate. While following the README to run testing, I encountered:

• Missing *_label.pkl under Real/test causing FileNotFoundError.
• Segmentation results results_test_*.pkl use pred_* keys, but the code expects class_ids/rois/scores/masks, likely causing a KeyError afterward.

Environment

• Ubuntu 20.04, Python 3.8.15, PyTorch 1.12.0, CUDA 11.3
• Repo: MonoDiff9D (latest main)
• Checkpoint: log1/rgb_diffusion_pose/epoch_80.pth
• Command:

python test.py --config config/rgb_diffusion_pose.yaml --test_epoch 80

Steps to Reproduce

1. Prepare datasets per README into data/ (including Real/test, obj_models/real_test.pkl, and segmentation_results/test_trainedwithMask, etc.).
2. Run the command above.

Actual Behavior

• Prints “no. of test images: 2754”, then crashes with:

FileNotFoundError: [Errno 2] No such file or directory: '.../data/Real/test/scene_1/0000_label.pkl'

• Also observed results_test_*.pkl keys are pred_class_ids/pred_bboxes/pred_scores/pred_mask, but later code uses class_ids/rois/scores/masks.

Questions / Requests

• Do we need to copy/link labels from the gts package into data/Real/test/scene_x/ to form *_label.pkl? Could the README include explicit instructions or a script?
• Can testing rely directly on gt_* fields already in results_*.pkl to avoid the *_label.pkl dependency?
• Could the code unify or support both naming styles (pred_* vs class_ids/rois/scores/masks), or provide an official conversion script?

谢谢！

[CNJianLiu]

Hi, pls make sure you have downloaded the complete datasets and put them in the correct directory

[luxilandu]

Summary

I’m a sophomore undergraduate. While following the README to run testing, I encountered:

• Missing *_label.pkl under Real/test causing FileNotFoundError.
• Segmentation results results_test_*.pkl use pred_* keys, but the code expects class_ids/rois/scores/masks, likely causing a KeyError afterward.

Environment

• Ubuntu 20.04, Python 3.8.15, PyTorch 1.12.0, CUDA 11.3
• Repo: MonoDiff9D (latest main)
• Checkpoint: log1/rgb_diffusion_pose/epoch_80.pth
• Command:

python test.py --config config/rgb_diffusion_pose.yaml --test_epoch 80

Steps to Reproduce

1. Prepare datasets per README into data/ (including Real/test, obj_models/real_test.pkl, and segmentation_results/test_trainedwithMask, etc.).
2. Run the command above.

Actual Behavior

• Prints “no. of test images: 2754”, then crashes with:

FileNotFoundError: [Errno 2] No such file or directory: '.../data/Real/test/scene_1/0000_label.pkl'

• Also observed results_test_*.pkl keys are pred_class_ids/pred_bboxes/pred_scores/pred_mask, but later code uses class_ids/rois/scores/masks.

Questions / Requests

• Do we need to copy/link labels from the gts package into data/Real/test/scene_x/ to form *_label.pkl? Could the README include explicit instructions or a script?
• Can testing rely directly on gt_* fields already in results_*.pkl to avoid the *_label.pkl dependency?
• Could the code unify or support both naming styles (pred_* vs class_ids/rois/scores/masks), or provide an official conversion script?

谢谢！

测试出来了，使用作者的210轮模型跑出来的结果跟论文贴出来的结果几乎一模一样

[shuaiqiqi1227]

Hello, how was this problem solved? I've encountered the same issue.