BCNN.py

import torch
import torch.nn as nn
eps = 1e-8

class BCNN(nn.Module):
    def __init__(self, thresh=1e-8, is_vec=True, input_dim=512):
        super(BCNN, self).__init__()
        self.thresh = thresh
        self.is_vec = is_vec
        self.output_dim = input_dim * input_dim

    def _bilinearpool(self, x, y=None):
        batchSize, dim, h, w = x.data.shape
        x = x.reshape(batchSize, dim, h * w)
        if y is not None:
            y = y.reshape(batchSize, dim, h * w)
            x = 1. / (h * w) * x.bmm(y.transpose(1, 2))
        else:
            x = 1. / (h * w) * x.bmm(x.transpose(1, 2))
        return x

    def _signed_sqrt(self, x):
        x = torch.mul(x.sign(), torch.sqrt(x.abs() + self.thresh))
        return x

    def _l2norm(self, x):
        x = nn.functional.normalize(x)
        return x

    def forward(self, x, y=None):
        if y is not None:
            x = self._bilinearpool(x,y)
        else:
            x = self._bilinearpool(x)
        x = self._signed_sqrt(x)
        if self.is_vec:
            x = x.view(x.size(0), -1)
        x = self._l2norm(x)
        return x