def build_multiscale_classifier(self, input_size):
n, c, h, w = input_size
hidden_shapes = []
for i in range(self.n_scale):
if i < self.n_scale - 1:
c *= 2 if self.factor_out else 4
h //= 2
w //= 2
hidden_shapes.append((n, c, h, w))
classification_heads = []
for i, hshape in enumerate(hidden_shapes):
classification_heads.append(
nn.Sequential(
nn.Conv2d(hshape[1], self.classification_hdim, 3, 1, 1),
layers.ActNorm2d(self.classification_hdim),
nn.ReLU(inplace=True),
nn.AdaptiveAvgPool2d((256, 1)),
))
self.classification_heads = nn.ModuleList(classification_heads)
self.logit_layer = nn.Linear(
self.classification_hdim * len(classification_heads),
self.n_classes)
self.logit_layer_gmm = nn.Sequential(
nn.AdaptiveAvgPool2d((self.classification_hdim, 1)),
nn.Linear(1, self.n_classes), nn.Softmax(dim=-1))
def _actnorm(size, fc):
if fc:
return FCWrapper(layers.ActNorm1d(size[0] * size[1] * size[2]))
else:
return layers.ActNorm2d(size[0])
