wei-mao-2019 / hisrepitself Goto Github PK

Hello, I would like to ask the difference between h36motion.py and h36motion3d.py in utils folder. Why didn't you end up using h36motion.py

Hi Wei,

I have a question about the way you generate the 3D keypoints xyz coordinates from the expmaps. As shown in this line this, all keypoints are calculated by the keypoint coordinate from the first frame p3d[0, i, :], which will generate errornous GTs(unless there is only 1 frame or batch_size=1, but in your dataloader this is not the case)

I think this line should be changed to

p3d[:, i, :] = torch.matmul(R[:, parent[i], :, :], p3d[:, i, :].unsqueeze(-1)).squeeze() + p3d[:, parent[i], :]

?

Or did I miss something?
Best

Hi,
Thank you for your great job!
For h36m on 3D xyz, I would like to ask why don't you train directly on all the 32 joints, but just use the 22 of them, please? (same for angle representations)
Thank you.

The in_feature in the code is determined by the dim_used parameter. In the code its size seems to be fixed 66, so the bug occured while ran this code with config python main_amass_3d.py --kernel_size 10 --dct_n 35 --input_n 50 --output_n 25 --skip_rate 5 --batch_size 128 --test_batch_size 128 --in_features 54 . Could you please explain how does in_feature work?

Hi,

in your paper you state that you are using the BMLrub dataset as test dataset. However in your code it looks to me like BioMotionLab_NTroje is hard coded for test. Could you provide some clarification here?

Also I was wondering if there is any particular reason why you are using your own forward kinematic instead of using the forward function of the Body Prior Model which returns the position for a given pose?

https://github.com/nghorbani/human_body_prior/blob/1936f38aec4bb959f6a8bf4ed304b6aafb42fa30/human_body_prior/body_model/body_model.py#L194

Thanks in advance!

作者您好，我想问一下在您在论文中使用的AMASS数据集只使用了18个关节点，这18个关节点相应的分布位置和编号能麻烦您介绍一下吗？

When I downloaded the entire human3.6m data set and tried to run the code, I encountered no such txt format file error. There are only cdf and MP4 format files in the human3.6m data set. May I ask how to convert the human3.6m data set How about converting the relevant dataset file format to txt format?

Hi,
Thank you for your great job!
Would you release your code about multi-level attention?

你好，请问现在数据集如何获取，斯坦福的那个数据库显示404

I run the command in the README file to test the pre-trained model, the error is

loading ckpt len from './checkpoint/pretrained/amass_3d_in50_out25_dctn30//ckpt_best.pth.tar'
Traceback (most recent call last):
File "main_amass_3d_eval.py", line 105, in
main(option)
File "main_amass_3d_eval.py", line 37, in main
net_pred.load_state_dict(ckpt['state_dict'])
File "/home/htang/anaconda3/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1051, in load_state_dict
raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
RuntimeError: Error(s) in loading state_dict for DataParallel:
size mismatch for module.gcn.gc1.weight: copying a param with shape torch.Size([60, 256]) from checkpoint, the shape in current model is torch.Size([70, 256]).
size mismatch for module.gcn.gc7.weight: copying a param with shape torch.Size([256, 60]) from checkpoint, the shape in current model is torch.Size([256, 70]).
size mismatch for module.gcn.gc7.bias: copying a param with shape torch.Size([60]) from checkpoint, the shape in current model is torch.Size([70]).

If the pre-train model was trained by setting the dct_n = 30?

How should I get AMASS data from this website