FasterRCNN Fine-tuning on Custom Dataset

I’m trying to fine-tune Faster RCNN model for object detection on KITTI dataset by following the PyTorch Mask RCNN tutorial. But when I try to train the model, I get ‘Loss is nan’ error immediately, and the training stops. Where shall I go from here?

1 Like

I guess there are probably many possible causes for that, but I had the same problem when I inadvertently fed this network images of range outside 0-1.