search

jiangtaoxie / fast-MPN-COV

@CVPR2018: Efficient unrolling iterative matrix square-root normalized ConvNets, implemented by PyTorch (and code of B-CNN,Compact bilinear pooling etc.) for training from scratch & finetuning.
http://peihuali.org/iSQRT-COV/index.html
MIT License
270 stars 56 forks source link

the loss tend to be nan #4

Closed ray-lee-94 closed 5 years ago

ray-lee-94 commented 5 years ago

when i used mpncovresnet50 and MPNCOV， the train converged. But if i change the backbone to resnet or VGG， keeping the MPNCOV unchanged, the train loss is nan. Beside, my dataset is for a mini-FGVC task. It contains 9 classes with extra-unbalanced. When i fine-tuning within two stage, The test acc is about 78, which is lower than plain vgg. Could you give me some advice? Thank you for your amazing work.

jiangtaoxie commented 5 years ago

@VCBE123 >change the backbone to resnet* or VGG*， keeping the MPNCOV unchanged, the train loss is nan.

There is a bug when reconstructing resnet, it's fixed. But for VGG, I did not observe this problem, can you show me your settings in train.sh?

>test acc is lower than plain vgg

Can you show me your experimental settings? Maybe need to adjust learning rate or weight decay.