About type error(triggers by datalodaer)

[JohnW2all]

Environment：

Dataset：

Using a testing command “python tools/test.py configs/das/exp_mupots.py weights/latest.pth --eval pck”, triggering an error:

[wangzt-halo]

pose_vis should be an array contained in the annotation file. Could you print out its value?

[JohnW2all]

pose_vis should be an array contained in the annotation file. Could you print out its value?

The value of pose_vis is "True", so the last sentence in the last picture says"TypeError: ' bool' object is not subscriptable".

Then I print the "ann['keypoints_vis']", it's an array full of "True".

Then I print the "ann_info".

I haven't changed the source code, so I'm confused about what went wrong.
I would like to ask if there are any errors in my coco dataset folder?

Thanks!

[wangzt-halo]

ok. you can try to replace the code if pose_vis[self.ROOT_IDX][0] == 0: with if pose_vis[self.ROOT_IDX] == 0: and see if this would work.

[JohnW2all]

ok. you can try to replace the code if pose_vis[self.ROOT_IDX][0] == 0: with if pose_vis[self.ROOT_IDX] == 0: and see if this would work.

I truly appreciate your timely help. It works!

After this a few small errors appear, now I am trying to fix it.

[JohnW2all]

ok. you can try to replace the code if pose_vis[self.ROOT_IDX][0] == 0: with if pose_vis[self.ROOT_IDX] == 0: and see if this would work.

Excuse me，I met a error about SVD and I don't know how to deal with this problem.

I printed the H and all the variables associated with it.

There are nan values among them. May I ask if this is the cause of the problem and could you please tell me how to fix it?
Thanks so much!

[wangzt-halo]

It seems that the problem comes from the wrong predictions. Are the outputs all `nan' for every data sample? If so, maybe there is something wrong with the trained model.

[JohnW2all]

Yes, all the output variable predict in the terminal contains nan.
I noticed that the terminal told me that the model and the weights did not match, which may have caused the problem.

Oh, I think I see the cause of the problem. Because I use the "latest.pth" provided by you, which is trained in the CMU Panoptic dataset.

And I use it incorrectly to predict in MuPoTS-3D dataset.

Thanks!

[tribeband]

I am not able to download the muco/human3.6m data from the link provided. will you please help me to provide another link?

[wangzt-halo]

@tribeband Because this data does not belong to us, we are sorry that we cannot provide you with another link.
But I checked the original link and it is still available for download.

[tribeband]

曾经我下载过，但是下完之后解压出错