I have a question about the Convolution 2D argument for the chain.
Defined in example/imagenet/nin.py
mlpconv1 = L.MLPConvolution2D(
3, (96, 96, 96), 11, stride = 4, wscale = w),
mlpconv2 = L.MLPConvolution2D (
96, (256, 256, 256), 5, pad=2, wscale=w),
mlpconv3 = L.MLPConvolution 2D (
256, (384, 384, 384), 3, pad = 1, wscale = w),
mlpconv4 = L.MLPConvolution 2D (
384, (1024, 1024, 1000), 3, pad=1, wscale=w),
In the network structure, who is the second argument array?
Also, what is the last wscale?
The first is the number of input channels, the third is the size of the convolution, and the fourth is the number of paddings.
The official document says number of output channels, but why array?
Professor, please.
python chainer machine-learning
It's not Convolution 2D, it's MLPConvolution 2D, so it's been extended.
http://docs.chainer.org/en/stable/_modules/chainer/links/connection/mlp_convolution_2d.html
wscale is the parameter used to determine the initial value of weights at random.
© 2024 OneMinuteCode. All rights reserved.