Closed Yuliang-Zou closed 5 years ago
I may do that in a few weeks. That said, as long as you have a GPU with >=16GB memory, you should be able to easily reproduce the results in README (within 1% for sure).
Thanks! So the current setting is for ResNet-101, right? I wonder if you have the hyperparameter settng for ResNet-152 as well.
DeepLab never tried to do ResNet-152, IIRC. So I am quite sure there is no "official" hyperparameters for ResNet-152. Personally I have never tried with ResNet-152.
I see. I will also try to play with this backbone and see if I can improve the result over ResNet-101.
Thanks for the great work! I wonder if any pre-trained model is available for download.