Add bias into SphereConv2D

deform_conv/deform_conv.py

@@ -117,7 +117,8 @@ def __init__(self,
  stride=1,
  padding=0,
  dilation=1,
- num_deformable_groups=1):
+ num_deformable_groups=1,
+ bias=True):
  super(ConvOffset2d, self).__init__()
  self.in_channels = in_channels
  self.out_channels = out_channels
@@ -129,6 +130,9 @@ def __init__(self,
 
  self.weight = nn.Parameter(
  torch.Tensor(out_channels, in_channels, *self.kernel_size))
+ self.bias = None
+ if bias:
+ self.bias = nn.Parameter(torch.zeros(1, out_channels, 1, 1))
 
  self.reset_parameters()
 
@@ -140,6 +144,9 @@ def reset_parameters(self):
  self.weight.data.uniform_(-stdv, stdv)
 
  def forward(self, input, offset):
- return conv_offset2d(input, offset, self.weight, self.stride,
- self.padding, self.dilation,
- self.num_deformable_groups)
+ output = conv_offset2d(input, offset, self.weight, self.stride,
+ self.padding, self.dilation,
+ self.num_deformable_groups)
+ if self.bias is not None:
+ output = output + self.bias
+ return output

sphere_cnn.py

@@ -9,7 +9,7 @@
 from deform_conv import ConvOffset2d
 
 
-# Calculate kernels of SphereCNN 
+# Calculate kernels of SphereCNN
 @lru_cache(None)
 def get_xy(delta_phi, delta_theta):
  return np.array([
@@ -32,7 +32,7 @@ def get_xy(delta_phi, delta_theta):
 
 @lru_cache(None)
 def cal_index(h, w, img_r, img_c):
- ''' 
+ '''
  Calculate Kernel Sampling Pattern
  only support 3x3 filter
  return 9 locations: (3, 3, 2)
@@ -105,19 +105,19 @@ def map_coordinates(input, coordinates, mode='bilinear', pad='wrap', slice_mode=
  coordinates.requires_grad = False
  h = input.shape[2]
  w = input.shape[3]
- 
+
  def _coordinates_pad_wrap(h, w, coordinates):
  coordinates[0] = coordinates[0] % h
  coordinates[1] = coordinates[1] % w
  return coordinates
- 
+
  def _coordinates_pad_zero(h, w, coordinates):
  out_of_bound_h = (coordinates[0] < 0) | (coordinates[0] > (h-1))
  out_of_bound_w = (coordinates[1] < 0) | (coordinates[1] > (w-1))
  coordinates[0, out_of_bound_h] = h
  coordinates[1, out_of_bound_w] = w
  return coordinates
- 
+
  if mode == 'nearest':
  coordinates = torch.round(coordinates).long()
  if pad == 'wrap':
@@ -154,12 +154,12 @@ class SphereConv2D(nn.Module):
  ''' SphereConv2D
  Note that this layer only support 3x3 filter
  '''
- def __init__(self, in_c, out_c, stride=1):
+ def __init__(self, in_c, out_c, stride=1, bias=True):
  super(SphereConv2D, self).__init__()
  self.stride = stride
- self.conv = ConvOffset2d(in_c, out_c, kernel_size=3, padding=1, num_deformable_groups=1)
+ self.conv = ConvOffset2d(in_c, out_c, kernel_size=3, padding=1, num_deformable_groups=1, bias=bias)
  self.offset = None
- 
+
  def forward(self, x):
  # x: (B, C, H, W)
  if self.offset is None or self.offset.shape[0] != x.shape[0] or\
@@ -186,7 +186,7 @@ def __init__(self, stride=1, mode='bilinear'):
  super(SphereMaxPool2D, self).__init__()
  self.mode = mode
  self.stride = stride
- 
+
  def forward(self, x):
  # x: (B, C, H, W)
  with torch.no_grad():
@@ -205,11 +205,11 @@ def forward(self, x):
  ridx = ridx[None, None, :, None]
  cidx = cidx[None, None, None, :]
  max_coord = coordinates[:, ridx, cidx, maxi]
- 
+
  return map_coordinates(x, max_coord, mode=self.mode, slice_mode='points')
 
- 
-if __name__ == '__main__': 
+
+if __name__ == '__main__':
  # test cnn
  cnn = SphereConv2D(3, 5, 1)
  out = cnn(torch.randn(2, 3, 10, 10))