John1231983

@Cpruce: I have not enough GPU to train coco. I just use the author pretrained coco. It provided 0.29 ap in seg task. Could you provide us your pretrain coco...

Thanks @lsy17096535 . Could you suggest any dataset and labelling tool to make landmark

@lsy17096535 : Thanks. Does hyperlandmark run on GPU or CPU? How to obtain the real-time landmark movement on video?

@tkuanlun350 : Have you try it? Thanks

FYI, DaLI is slower than pytorch in distributed training. I tried it

I have tried mobinetv3 with small configure but it does not work. Any reason?

Thanks @cavalleria but the log is unnormal. ``` Training Prec@1 44.824 (45.528) Training Prec@5 58.203 (57.886) ``` It should be 99%

@xsacha @cavalleria thanks your comments but I refer log from https://github.com/HuangYG123/CurricularFace The log shows at the same epoch ```Training Prec@1 99.8 (100.0) Training Prec@5 100 (100)```

Sorry it is typo :)

Very good discussion for data augmentation. I have some questions after reading the thread 1. Does motion blur and hflip improve performance? Other augmentation may decrease accuracy on normal face...