HigherHRNet-Human-Pose-Estimation icon indicating copy to clipboard operation
HigherHRNet-Human-Pose-Estimation copied to clipboard

Published 20 hours ago • verified publisher icon HRNet

Network fails to learn with change in input resolution

Open anerisheth19 opened this issue 4 years ago • 1 comments

I am using your code base with a variation in input resolution from 512x512 to 1024x576. I changed the transforms, coco_dataset, loss functions and made it work for rectangular images. However, after 30 epochs, when I evaluate, the AP results are terrible.

Can you throw some insights on why your code works bad for rectangular images?

anerisheth19 avatar Jul 31 '20 18:07 anerisheth19

I meet the same problem, have you solve it?

wusaisa avatar Nov 12 '23 10:11 wusaisa