expose
expose copied to clipboard
vchoutas
Fail to crop the dataset
Hi, this is really nice work.However, when i tried to use the dataset and try to crop the data, i got the error like this "ValueError: could not broadcast input array from shape (57,97,3) into shape (57,25,3)". Seems like the annotations are not properly align to the exact image. Or maybe do i miss something? Thanks.
I used the CuratedFittings dataset. I just want how to preprocess the dataset, but suddenly I got that error. Based on my experiment, when i tried to crop lsp extension dataset, i got that error not all images only some images. Do you have any idea?
And i have a question about the evaluation metric (not related with this issue). In table 1 use compare with 3DPW in term of V2V. But expose output is SMPLX which is around 10475 vertices. but SMPL is 6890. Maybe you have mapping vertices index from SMPLX to SMPL. Maybe can you share the mapping? or can you give the alternative way how to compute v2v with 3DPW dataset. Thanks!
@joshuajano We released code in the smplx repository to transfer SMPL to SMPL-X parameters. That's what we used to evaluate for evaluate to get SMPL-X V2V numbers.
Can you share the command that you used? Or do you have the indices of the images that failed?
Sorry, So these are examples of fail cases (Randomly select) /lsp/lspet/images/im02007.png /lsp/lspet/images/im07409.png /lsp/lspet/images/im02812.png /lsp/lspet/images/im04134.png /lsp/lspet/images/im07178.png This error occurs when I run cropping function https://github.com/vchoutas/expose/blob/dffc38d62ad3817481d15fe509a93c2bb606cb8b/expose/utils/transf_utils.py#L74. But it just happen on some images on lspet dataset.
Do you have the training code for this paper? Can you share it with me? This is my email address [email protected]，Thank you very much.！
