Accuracy issue on VGG

I am trying to train VGG on mxnet with 16 layer and lr=0.05 but after 50 epch also it is not converging (accuracy 0.01) .

which dataset are you training on?

1 Like

@olivcruche I am using imagenet dataset

1 Like

Can you post a snippet of your training code.

Better yet if you’re just trying to train VGG16 on imagenet. You can go to

https://gluon-cv.mxnet.io/model_zoo/classification.html

and click on Download train_imagenet.py

Once you have the script downloaded, then you can run:

https://raw.githubusercontent.com/dmlc/web-data/master/gluoncv/logs/classification/imagenet/vgg16.sh

Hope that helps.

2 Likes

@sad thanks for help. After enabling batchnorm it is converging

1 Like