poor inference. about prnet_pytorch HOT 5 CLOSED

what exactly is this variable 'self.uv_kpt_ind'?

from prnet_pytorch.

The point is somewhat related to the model itself. I think you can train prnet with more data.

from prnet_pytorch.

uv_kpt_ind

official landmark position of shape 2 x 68

from prnet_pytorch.

did you try to reproduce the gifs in the readme? they seem to be spot on while my output seems to be total garbage.

Also, so you're telling me my code snippet is not logically erroneous, right?

official landmark position of shape 2 x 68

so the net is using this to infer on all test data? I'll read up on this, don't exactly understand how this is working.

from prnet_pytorch.

did you try to reproduce the gifs in the readme? they seem to be spot on while my output seems to be total garbage.

Also, so you're telling me my code snippet is not logically erroneous, right?

official landmark position of shape 2 x 68

so the net is using this to infer on all test data? I'll read up on this, don't exactly understand how this is working.

Yep, the inference code you posted is the same as me. I recommend you transform all the data psoted as in the original PRNet repo (not only 300WLP) and train with it.

from prnet_pytorch.