[Caffe2] Does python API support ReduceMean?

JingyuQian · August 28, 2018, 10:41pm

Hi,
I’m trying to implement the same structure described in this squeeze_excitation paper: https://arxiv.org/pdf/1709.01507.pdf. The first layer of the structure is a global pooling layer across height and width channels. I searched Caffe2 python API to found a suitable operator to do this but didn’t found any.
There is a ReduceMean op, however, written for C++ API. It calculates the mean value across the last dimension of the input blob.
Does anyone know how I can make this work? Thanks in advance!

Richard_Pruss · October 31, 2018, 7:57pm

You looking for:
https://pytorch.org/docs/stable/nn.html#adaptiveavgpool2d

You can see it in this implementation of that paper:

github.com

moskomule/senet.pytorch/blob/master/se_module.py

from torch import nn


class SELayer(nn.Module):
    def __init__(self, channel, reduction=16):
        super(SELayer, self).__init__()
        self.avg_pool = nn.AdaptiveAvgPool2d(1)
        self.fc = nn.Sequential(
                nn.Linear(channel, channel // reduction),
                nn.ReLU(inplace=True),
                nn.Linear(channel // reduction, channel),
                nn.Sigmoid()
        )

    def forward(self, x):
        b, c, _, _ = x.size()
        y = self.avg_pool(x).view(b, c)
        y = self.fc(y).view(b, c, 1, 1)
        return x * y