About forward about prnet_pytorch HOT 9 CLOSED

Hi, I think a np.clip(uv_map, 0, 1) should be added here to fix this problem?

from prnet_pytorch.

Hi, I have inference stage finished and upload my test model results/latest.pth, you can test it!

However, I found there may be some mistakes in the original face3d repo. I will strive to fix that in the coming days.

from prnet_pytorch.

Hi, I have inference stage finished and upload my test model results/latest.pth, you can test it!
However, I found there may be some mistakes in the original face3d repo. I will strive to fix that in the coming days.

The author don't say how to normalize pos_map label, when the output of resfcn256 is in [0, 1] range and *.npy have some negative elements.

from prnet_pytorch.

Hi, I have inference stage finished and upload my test model results/latest.pth, you can test it!
However, I found there may be some mistakes in the original face3d repo. I will strive to fix that in the coming days.

The author don't say how to normalize pos_map label, when the output of resfcn256 is in [0, 1] range and *.npy have some negative elements.

Hi, I just follow the process as same as the original repo. Do you mean that there may be some problems about normalization in face3d ?

from prnet_pytorch.

Hi, I have inference stage finished and upload my test model results/latest.pth, you can test it!
However, I found there may be some mistakes in the original face3d repo. I will strive to fix that in the coming days.

The author don't say how to normalize pos_map label, when the output of resfcn256 is in [0, 1] range and *.npy have some negative elements.

Hi, I just follow the process as same as the original repo. Do you mean that there may be some problems about normalization in face3d ?

I think problems are in traning part, because *.npy labels have some negative elements, but the output of resfcn256 is Sigmoid which output range is [0, 1]. Your labels:
only devided by 255.0? But I don't know how to normalize the labels by the way @@!

from prnet_pytorch.

This is my final results.

from prnet_pytorch.

This is my final results.

Looks good to me, good job~

from prnet_pytorch.

This is my final results.

Hi, I want to know how you can get such a good result. I can train and forward, but the result is poor.

from prnet_pytorch.

This is my final results.

Hi, I want to know how you can get such a good result. I can train and forward, but the result is poor.

Based on my own view, you simply need more training data to make your network more robust. 300WLP provided by me is not enough for a precise high-varied viewpoint adaptive model.

from prnet_pytorch.