Open GeoffreyChen777 opened 6 years ago
I think one of the problem might be that all of the three parameters (images, bboxes, bb_index) should be on the same devices. I resolve my illegal memory access problem by moving bb_index to the cuda devices by bb_index.to(torch.device("cuda" if torch.cuda.is_available() else "cpu"))
there might be something wrong with your pretrained model when you copyed it to /data/pretrained_model , you'd better make sure that your pretrained model is complete and without broken in the path
Checking the image crop size may be helpful.
I use your Roialign in my code, and I got this error, RuntimeError: cuda runtime error (77) : an illegal memory access was encountered at torch/csrc/cuda/Module.cpp:267
I tried the roialign in pytorch-faster-rcnn and got no error.
Thank you.