liuzhuang13 / DenseNet

Densely Connected Convolutional Networks, In CVPR 2017 (Best Paper Award).
BSD 3-Clause "New" or "Revised" License
4.73k stars 1.07k forks source link

Wide-DenseNet #20

Closed 1292765944 closed 7 years ago

1292765944 commented 7 years ago

Congrats for best paper on CVPR 2017! I'm troubled with the memory problem with densenet. Would you share your wide-densenet implementation and pre-train models publicly?

Best!

liuzhuang13 commented 7 years ago

Hello @1292765944 and thanks!

To run a wide-densenet just set -depth to be relatively small (e.g. 40) and -growthRate to be relatively large (e.g. 48), there's no seperate implementation for wide-densenet.

And sorry we didn't train any wide-densenets on ImageNet, so no pre-trained models are available now.

1292765944 commented 7 years ago

@liuzhuang13 You mean DenseNet-161 (k=48)? I notice that your imagenet pre-trained models all have large growthRate(32 or 48) . So wide-densenet means more wider?

liuzhuang13 commented 7 years ago

@1292765944 Yes, probably a larger growth rate will do better, since ImageNet is such a complex dataset.