pytorch-faster-rcnn
pytorch-faster-rcnn copied to clipboard
RPN Returning Nan for Kitti Data
Hi,
Seeing the following error: https://stackoverflow.com/questions/57698277/getting-nan-result-out-of-resnet101-backbone-with-kitti-images
When trying to run Kitti in testing mode. Will try using different weights to see if that fixes the issue. Has anyone run into this? With Kitti or not?
Figured it out, pascal VOC pixel location index starts at 1, but kitti starts at 0. need to remove all -1's from BBox generation.