correct dimension for testing on custom dataset

[maryamag85]

I am trying to run the try on module on my own dataset however I am not sure about the correct dimension of the input images and masks etc. what should be the target and also dataset dimnesion in demo file (ds = Dataset(dataroot=dataroot, dim=(?,?), n_human_part=8)

with my own dimensions that I set org_size = 750,110, like sample images and target_size=256,176 as the template is I get following output on my own images Please help me figure out the issue It would be great if you provide a more detailed manual for preparing the custom dataset. like if resizing is needed and how to bring keypoints from 25 to 18. I have followed and read all the issues and I saw the template you provided for loading from pose.json file however even if I load from json I still need to set kepoints in the standard_test_anns.txt. it is confusing how to get the correct 18 keypoints out of 25 and how to set that file . your input is appreciated.

Thanks

[cuiaiyu]

Can you visualize the keypoints?

[maryamag85]

the keypoints are visualized correctly on the custom image. My confusion is about how to set the image dimensions. when I visualize the sample images keypoints in the provided dataset , keypoints are in the 176*256 dimension however original set are 750 * 1101 dimension. Is it possible for you to provide a clear instruction for data preparation to avoid confusion. my dimensions are the issue I guess.

[netsalesbg-ltd]

@maryamag85 I suggest first resizing the original image to 176x256 and then running open pose on COCO model to get the correct keypoints.