Closed PearlDzzz closed 7 years ago
The contribution of squeeze net is the model size not the computational complexity~~ Actually, computational complexity of Alexnet and SqueezeNet are similar~
Interesting issue! @forresti would you please explain why much less parameters but use much more time?
@maydaygmail the most majority of model parameter is in FC layers. and the most computation cost at conv layers. Squeezenet use more Conv layers, it's much deeper than alexnet
Thanks @austingg. I think there is a implementation problem. In fire module, expand layer has 2 different convolution, and the 2 different convolution are calculating serially not parallel
@macd @forresti @antingshen @samster25 @terrychenism Hi, I use command" caffe.exe time --model=SqueezeNet_v1.1_deploy.prototxt -gpu 0 -iterations 100 " to test the time. AlexNet:11ms, SqueezeNet:30ms. even use cudnn v4, the time of SqueezeNet is still twice or even three times than AlexNet. Do you have any advice?