Open ApoorvaSuresh opened 4 years ago
i have tried to augment the depth channel as heatmaps.
But I always get nan loss for rpn_bbox_loss. can someone help?
This repository that extends Mask R-CNN for RGB-D data might be of interest to you: https://github.com/orestis-z/mask-rcnn-rgbd. I includes also image augmentation for all 4 channels: https://github.com/orestis-z/mask-rcnn-rgbd/blob/d590e0f5085f8cbe895a6698e284426fd0116aa4/model.py#L1225-L1277
Hi, I am training this model with my own data- RGBD. But I am unable to apply image augmentation. Do the image augmentors works only on 3 channel images?