Hang Zhang

Results 279 comments of Hang Zhang

> Maybe we can do local distributed training style instead of multithreading The SyncBN is implemented in operator level, which does not support distributed training. The synchronization happens within the...

The label -1 is ignored during the training and evaluation.

Thanks for the comments! I agree and will add more comments about the util functions.

Link to model unitest: https://github.com/dmlc/gluon-cv/blob/master/tests/unittests/test_model_zoo.py#L147

> Like @chinakook said please try to remove usage of `.shape` in the network. @sufeidechabei I think you may allow an argument during the init function to fix the input...

For detection models, we have FPN, Mask RCNN, YOLO V3, RetinaNet in our roadmap. @zhreshold please corrects me, if it is wrong.

Cityscapes requires larger crop size, which is very memory consuming. Unfortunately, single GPU won't work

ResNeSt has the same interface as the ResNet. You should be able to find some example of doing that.

ImageNet training is here https://github.com/zhanghang1989/ResNeSt/tree/master/scripts/gluon