Open alushnikov opened 6 years ago
Hi,
thanks for your interest : )
It is a known issue that DenseNet takes a lot of GPU memory. This issue has been investigated in this paper and you can find the corresponding code in here. Unfortunately, this method has not been implemented using TensorFlow yet.
Due to other agenda, I am not going to implement this method in the near future. But it is definitely worth a shot.
Hi!
Thanks for your project, very nice implementation! Plugs into tf.slim seamlessly!
It'd be great if it could be used for bigger images/datasets like ImageNet. Did you try running your densenet implementation for larger L,k values like described in paper?
I have GTX 1080Ti, but run into OOM Error even with {L=100, k=12}. I might be looking to adjust this implementation to be memory efficient. Just wanted to check if you tried doing this already.