generate pose coordinates as output about corrpm

[ziwei-zh]

Hi, you can select the max value along every channel, like pose1.max(1)

[lizc126]

Hi, you can select the max value along every channel, like pose1.max(1)

Hi, thank you for your reply! So i converted 96x96 back to 384x384 the same way you did for seg2, and selected the pixels with the max values for each of the 16 channels. However, i noticed that 1) for each channel, there are several pixels with the same max values, althouth they are pixels nearby each other 2)when I tried to map the 16 keypoints on the 384x384 test image, the mapping seemed wrong. Is there any steps I am missing? Thank you!

[ziwei-zh]

1. First you may have to do NMS or other post-processing. You may refer to simple baseline repo: https://github.com/microsoft/human-pose-estimation.pytorch/blob/18f1d0fa5b5db7fe08de640610f3fdbdbed8fb2f/lib/core/inference.py#L49
2. I think the upsampling step is not needed since it may cause misalignment issues.

[lizc126]

1. First you may have to do NMS or other post-processing. You may refer to simple baseline repo: https://github.com/microsoft/human-pose-estimation.pytorch/blob/18f1d0fa5b5db7fe08de640610f3fdbdbed8fb2f/lib/core/inference.py#L49
2. I think the upsampling step is not needed since it may cause misalignment issues.

Thank you very much! Finally able to get it work;)

[cLEARLoVE00]

Hello,Can you share test code to output points?I also want to output 16 points @Sonseca97 ,Thanks