search

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

Covolution before entering the first dense block for imagenet dataset #46

Open anthony123 opened 6 years ago

anthony123 commented 6 years ago

Hi, there

For the imagenet dataset, DenseNet use 7x7 Conv before entering the first dense block; I also read the following paper condensenet , which use 3x3 conv before entering the first block. I wonder if i can change the 7x7 conv to 3x3, and keep the pooling unchanged (since it makes densenet more parameter- efficient). Does it hurt DenseNet's performance on imagenet?

gaohuang commented 6 years ago

Replacing the 7x7 conv layer with a couple of 3x3 conv layers may slightly improve the performance, if you keep the number of parameters roughly unchanged. That's why CondenseNet adopts that design choice.