egeozsoy / 4d-or Goto Github PK

Hi, I try to infer on the test set using 3DSSG, followed by "python -m scene_graph_prediction.main --config no_gt_image.json". But an error is displayed:
Original Traceback (most recent call last):
File "/home/allen/anaconda3/envs/4dor/lib/python3.7/site-packages/torch/utils/data/_utils/worker.py", line 287, in _worker_loop
data = fetcher.fetch(index)
File "/home/allen/anaconda3/envs/4dor/lib/python3.7/site-packages/torch/utils/data/_utils/fetch.py", line 49, in fetch
data = [self.dataset[idx] for idx in possibly_batched_index]
File "/home/allen/anaconda3/envs/4dor/lib/python3.7/site-packages/torch/utils/data/_utils/fetch.py", line 49, in
data = [self.dataset[idx] for idx in possibly_batched_index]
File "/home/allen/MI-projects/4D-OR/scene_graph_prediction/scene_graph_helpers/dataset/or_dataset.py", line 88, in getitem
human_name_to_3D_joints = np.load(str(OR_4D_DATA_ROOT_PATH / 'human_name_to_3D_joints' / f'{take_idx}_GT_True.npz'), allow_pickle=True)[
File "/home/allen/anaconda3/envs/4dor/lib/python3.7/site-packages/numpy/lib/npyio.py", line 417, in load
fid = stack.enter_context(open(os_fspath(file), "rb"))
FileNotFoundError: [Errno 2] No such file or directory: '/4D-OR/human_name_to_3D_joints/2_GT_True.npz'

I checked the dataset and found that there is indeed no ‘2_GT_True.npz' and '6_GT_True.npz'. how can I get the result of 'infer'? Thank you.

Hello, I am wondering how I can use a unified scale to locate the position of both humans and objects such as "anesthesia_equipment", I use annotations in export_holistic_take_processed folder as human position points and use GroupFree3D dataset's "center_label" as object's position, but it seems that they are in different scales or processed with different standard methods. How can I get the unified scaled position of both human and object? Thank you.

The title says it all.

Hi, I have a question about the track id. I saw that the track id can go up to almost 20 in each sequence, but obviously there isn't that much of person according to the human name. Therefore, may I ask how did you define the track id? And do you recommend using the human name or the track id to conduct the tracking task? Thank you!

Hi, thank you for your great work, but I am facing a problem about the accuracy of the 3d human pose labels.

I was running the 3d human pose estimation experiment based on VoxelPose, but according to the visualization of the projected 2d views ('image_with_joints' folder), the gt labels are often inaccurate. I am attaching some of the visualization results.

I checked the code, but I don't see any problems. So I want to ask that does this kind of inaccuracy meet the expectations? Because I don't know the details of the annotation process.

Thank you!

Hi, how to evaluate the results of the clinical role predictions of my model on the 4D-OR test set?