hello, thank you for your awesome work. I have toubles of using multi-gpus when training:

I added "model = nn.DataParallel(model)" before main.py line187: "all_param = []", but it doesn't work and gives an error: Traceback (most recent call last): File "main.py", line 190, in for i_model in model: TypeError: 'DataParallel' object is not iterable

can you please tell me how to solve this question? thank you!

The code for testing uses the GT 3D as well as camera parameters in the refine step. I got the normal step working, but with a wild input, what should I feed into the refine network?

When try to python main.py --refine --lr 1e-5 --reload --previous_dir [your model saved path], the loss is always 0 overall process, is it normal? I can't get the same result in the paper 《Exploiting Temporal Contexts with Strided Transformer for 3D Human Pose Estimation》.

is it refer to Exploiting Spatial-temporal Relationships for 3D Pose Estimation via Graph Convolutional Networks, the refinement module? how many epochs do I need to train to add this module？

Thank you for providing the inference code. I am very interested in your paper. According to the paper and code, it seems like using 351 frames apparently. I tried to modify vis.py code to use 27 frames. Unfortunately I faced a hardship. The following is the error code that I got. Could you give me a instruction?

size mismatch for Transformer.pos_embedding: copying a param with shape torch.Size([1, 351, 256]) from checkpoint, the shape in current model is torch.Size([1, 27, 256]). size mismatch for Transformer_reduce.model.pos_embedding_1: copying a param with shape torch.Size([1, 351, 256]) from checkpoint, the shape in current model is torch.Size([1, 27, 256]). size mismatch for Transformer_reduce.model.pos_embedding_2: copying a param with shape torch.Size([1, 351, 256]) from checkpoint, the shape in current model is torch.Size([1, 27, 256]). size mismatch for Transformer_reduce.model.pos_embedding_3: copying a param with shape torch.Size([1, 351, 256]) from checkpoint, the shape in current model is torch.Size([1, 27, 256]).

Dear author,

Thank you for your amazing work. I trying run your demo code with in-the-wild video and I got only about 15fps with 3D pose estimation (with HRnet and yolo detection) on my RTX3090 GPU. I checked gpu usage and it use only about 2GB of my memory. I want to ask that is there any setting that limmit my gpu usage and anyway to increase fps when running demo? I have install CUDA 11.x already. According to the paper, the method can achieve real-time application. Am I doing wrong some where? sorry, I'm new to this deeplearning field. Thank you very much for your time!

How did you deal with the situation when the target frame is the first one of the video? then there isn't any perceding frames to make the target frame as the 'center frame'. Or you just ignore it and start the evaluation from the 13th frame when the input sequence length is 27?

When I train the model using python main.py , it works well. But when I try to refine the model, using python main.py --refine --lr 1e-5 --reload --previous_dir it reports error：

Traceback (most recent call last): File "main.py", line 225, in loss = train(opt, actions, train_dataloader, model, optimizer_all, epoch) File "main.py", line 23, in train return step('train', opt, actions, train_loader, model, optimizer, epoch) File "main.py", line 94, in step loss.backward() File "D:\Anaconda\envs\pose\lib\site-packages\torch_tensor.py", line 307, in backward torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs) File "D:\Anaconda\envs\pose\lib\site-packages\torch\autograd_init_.py", line 156, in backward allow_unreachable=True, accumulate_grad=True) # allow_unreachable flag RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.FloatTensor [64, 1024]], which is output 0 of ReluBackward0, is at version 1; expected version 0 instead. Hint: the backtrace further above shows the operation that failed to compute its gradient. The variable in question was changed in there or anywhere later. Good luck!

I can't fine the problem, I almostly use with torch.autograd.set_detect_anomaly(True): , but still can't find the crux.

After training the model according to README, I typed python demo/vis.py --video back.webm and got the following error.

(venv) yuki@take-ubuntu2:~/workspace/StridedTransformer-Pose3D$ python demo/vis.py --video back.webm

Generating 2D pose...
100%|███████████████████████████████████████████████████████████████████████████| 499/499 [00:44<00:00, 11.21it/s]
Generating 2D pose successful!

Generating 3D pose...
  0%|                                                                                     | 0/499 [00:00<?, ?it/s]
Traceback (most recent call last):
  File "demo/vis.py", line 276, in <module>
    get_pose3D(video_path, output_dir)
  File "demo/vis.py", line 222, in get_pose3D
    show3Dpose( post_out, ax)
  File "demo/vis.py", line 68, in show3Dpose
    ax.set_aspect('equal')
  File "/home/yuki/workspace/StridedTransformer-Pose3D/venv/lib/python3.6/site-packages/mpl_toolkits/mplot3d/axes3d.py", line 324, in set_aspect
    "Axes3D currently only supports the aspect argument "
NotImplementedError: Axes3D currently only supports the aspect argument 'auto'. You passed in 'equal'.

I looked into this error and found that the following line seems to be the cause.

https://github.com/Vegetebird/StridedTransformer-Pose3D/blob/5fbb0248ce6dd7db51f669422ed2dc86556e3f7a/demo/vis.py#L68

I replaced this line with ax.set_aspect('auto') or ax.set_box_aspect([1,1,1]) according to https://github.com/fabro66/GAST-Net-3DPoseEstimation/issues/51 and it works fine.

Why do you write ax.set_aspect('equal')?

If there is no particular reason then it needs to be fixed.

Hello, I am very interested in your work, but I have not been successful in trying to use a general visualization method. Could you please provide the visualization code?

Wonderful work! I just noticed this teeny-tiny error, in main.py < line 175< refine_path object initialize as a list (While it needn't be). This causes the following error:

So, either line 175 should be removed or line 178 should become "refine_path.append(path)", which may cause further downstream errors due to refine_path being a list. Please check that out!

Thanks!