def forward(self, x, out_feat_keys=None):
"""Forward an image `x` through the network and return the asked output features.
Args:
x: input image.
out_feat_keys: a list/tuple with the feature names of the features
that the function should return. By default the last feature of
the network is returned.
Return:
out_feats: If multiple output features were asked then `out_feats`
is a list with the asked output features placed in the same
order as in `out_feat_keys`. If a single output feature was
asked then `out_feats` is that output feature (and not a list).
"""
out_feat_keys, max_out_feat = parse_out_keys_arg(
out_feat_keys, self.all_feat_names)
out_feats = [None] * len(out_feat_keys)
feat = x
for f in range(max_out_feat + 1):
feat = self._feature_blocks[f](feat)
key = self.all_feat_names[f]
if key in out_feat_keys:
out_feats[out_feat_keys.index(key)] = feat
return out_feats
def forward(self, x, out_feat_keys=None):
out_feat_keys, max_out_feat = parse_out_keys_arg(
out_feat_keys, self.all_feat_names)
out_feats = [None] * len(out_feat_keys)
feat = x
for f in range(max_out_feat + 1):
feat = self._feature_blocks[f](feat)
key = self.all_feat_names[f]
if key in out_feat_keys:
out_feats[out_feat_keys.index(key)] = feat
return out_feats
def forward(self, x, out_feat_keys=None):
"""Forward an image `x` through the network and return the asked output features.
Args:
x: input image.
out_feat_keys: a list/tuple with the feature names of the features
that the function should return. By default the last feature of
the network is returned.
Return:
out_feats: If multiple output features were asked then `out_feats`
is a list with the asked output features placed in the same
order as in `out_feat_keys`. If a single output feature was
asked then `out_feats` is that output feature (and not a list).
"""
out_feat_keys, max_out_feat = parse_out_keys_arg(
out_feat_keys, self.all_feat_names)
# pool5, 12
out_feats = [[]] * len(out_feat_keys)
x_size = x.size()
if len(x_size) == 5:
#unsupervise learning
B, T, C, H, W = x_size
x = x.transpose(0, 1)
for i in range(T):
feat = x[i]
for f in range(max_out_feat + 1):
feat = self._feature_blocks[f](feat)
key = self.all_feat_names[f]
if key in out_feat_keys:
out_feats[out_feat_keys.index(key)].append(feat)
for key in out_feat_keys:
out_feats[out_feat_keys.index(key)] = torch.cat(
out_feats[out_feat_keys.index(key)], 1)
return out_feats
else:
feat = x
for f in range(max_out_feat + 1):
feat = self._feature_blocks[f](feat)
key = self.all_feat_names[f]
if key in out_feat_keys:
out_feats[out_feat_keys.index(key)] = feat
return out_feats
def forward(self, x, out_feat_keys=None):
out_feat_keys, max_out_feat = parse_out_keys_arg(
out_feat_keys, self.all_feat_names)
out_feats = [[]] * len(out_feat_keys)
x_size = x.size()
if len(x_size) == 5:
#unsupervise learning
B, T, C, H, W = x_size
x = x.transpose(0, 1)
for i in range(T):
feat = x[i]
for f in range(max_out_feat + 1):
feat = self._feature_blocks[f](feat)
#print(feat.size())
key = self.all_feat_names[f]
if key in out_feat_keys:
out_feats[out_feat_keys.index(key)].append(feat)
for key in out_feat_keys:
out_feats[out_feat_keys.index(key)] = torch.cat(
out_feats[out_feat_keys.index(key)], 1)
print(out_feats[0].size())
return out_feats
else:
feat = x
for f in range(max_out_feat + 1):
feat = self._feature_blocks[f](feat)
key = self.all_feat_names[f]
if key in out_feat_keys:
out_feats[out_feat_keys.index(key)] = feat
return out_feats
def forward(self, x, out_feat_keys=None):
"""
Override forward function of standard AlexNet
"""
feat = x
if cfg.MODEL.FEATURE_EVAL_MODE:
# Evaluation of SSL features
out_feat_keys, max_out_feat = parse_out_keys_arg(
out_feat_keys, self.all_feat_names)
out_feats = [None] * len(out_feat_keys)
for f in range(max_out_feat + 1):
feat = self._feature_blocks[f](feat)
key = self.all_feat_names[f]
if key in out_feat_keys:
out_feats[out_feat_keys.index(key)] = feat
return out_feats
else:
# Pretext Training: process each puzzle piece through model
batch_dim, jigsaw_dim, _, _, _ = feat.size()
feat = feat.transpose(0, 1)
output = []
for i in range(9):
jigsaw_piece = feat[i]
for layer in self._feature_blocks:
jigsaw_piece = layer(jigsaw_piece)
jigsaw_piece = self.fc6(jigsaw_piece)
output.append(jigsaw_piece)
output = torch.cat(output, dim=1)
return [output]
