# detectron2:bug record ### image is bluish https://blog.csdn.net/u013608424/article/details/80117178 What image cv2 read is construct in BGR,while plt.imgshow() will read the image as RGB, so the picture will become bluish ### RuntimeError: cuda runtime error (710) : device-side assert triggered at /pytorch/aten/src/THC/generic/THCTensorMath.cu:196 https://discuss.pytorch.org/t/runtimeerror-cuda-runtime-error-59-device-side-assert-triggered-at-pytorch-aten-src-thc-generic-thctensormath-cu-24/32438 I forget to change the output channel ### AttributeError: Cannot find field 'gt_masks' in the given Instances! https://github.com/facebookresearch/detectron2/issues/485 Bug is caused by using mask RCNN model to detect the data without mask label ### Get -1 in AP value I get -1 in all the AP value, but I check the image and that trained well. I found that I set wrong in bounding box. It can show the box in correct palce and train to get right place, but can't evaluate the AP value. Moreover, you have to delete the .json file in output to evaluate the new model. ###### tags: `FasterRCNN` `LAB`