Giters

Lotayou / everybody_dance_now_pytorch

A PyTorch Implementation of "Everybody Dance Now" from Berkeley AI lab.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

size compatibility issues

dklingmann opened this issue · comments

commented

There seem to be compatibility issues of image sizes between README.md and pose_estimator/compute_coordinates_for_video.py . README.md suggests the frames should be resized into 288*512 (for 16:9 HD video) but this leads to an issue when using the function 'cordinates_from_image_file' from compute_coordinates_for_video.py.

Specifically, the imageToTest_padded, line 189, will not be the right size for the model suggested in the READ.me file which takes images of size (368, 368). This doesn't work with the given scale_search (line 25) and therefore multiplier (line 179).

A workaround is to change the transformation (line 189) in the function cordinates_from_image_file to ensure the input into the model has the right size.