Kwangyeon Gill comments

Results 6 comments of


                                            Kwangyeon Gill

Ground-truth bounding boxes included in RPN proposals?

See https://github.com/facebookresearch/maskrcnn-benchmark/issues/570#issuecomment-473218934. Basically, it's a hack to train the classifier, without the box regressor obtaining the gradients.

`TORCH_LIBRARY` and `m.def` Not Working as Documented

https://github.com/andylizf/extension-cpp/blob/2d49e184f82ab6e1b61e8dc3abb6c7ede65ca37b/extension_cpp/csrc/lltm.cpp#L9 I think it should be `TORCH_LIBRARY(extension_cpp, m)`, not `TORCH_LIBRARY(TORCH_EXTENSION_NAME, m)`

Weird bug when setting `val_check_interval` dynamically in `setup()`

Just found another internal variable `val_check_batch` in `FitLoop`, and changing its value and `val_check_interval` to the dynamically calculated value does the job. This behavior is still weird, as assigning a...

Torchvision Faster R-CNN onnx export with dynamic batch size fails during inference

> Hi @davidgill97 , sorry, I don't think I'll be able to prioritize ONNX-related issues from now. I see. I'm willing to look into the problem, could you give me...

OSError: resource or device busy

Can you check setting `persistent_workers=True`?

No backend type associated with device type cpu

I've looked into the problem and found out that the main reason for this error is that default distributed backend for lightning is nccl. If `compute_on_cpu=True`, it gives the error...