Fixed calculation of convolution output shape (8a1ad36d) · Commits · 钟慕尧 / deepchem

deepchem/models/tensorgraph/layers.py

+31 −21

Original line number	Diff line number	Diff line
		@@ -446,6 +446,16 @@ class SharedVariableScope(Layer):
		return self._shared_with._get_scope_name()


		def _conv_size(width, size, stride, padding):
		"""Compute the output size of a convolutional layer."""
		if padding.lower() == 'valid':
		return 1 + (width - size) // stride
		elif padding.lower() == 'same':
		return 1 + (width - 1) // stride
		else:
		raise ValueError('Unknown padding type: %s' % padding)


		class Conv1D(Layer):
		"""A 1D convolution on the input.

		@@ -544,11 +554,11 @@ class Conv1D(Layer):
		parent_shape = self.in_layers[0].shape
		if isinstance(strides, int):
		strides = (strides,)
		if padding.lower() == 'same':
		if isinstance(kernel_size, int):
		kernel_size = (kernel_size,)
		self._shape = (parent_shape[0],
		int(np.ceil(parent_shape[1] / strides[0])), filters)
		else:
		self._shape = (parent_shape[0], parent_shape[1] // strides[0], filters)
		_conv_size(parent_shape[1], kernel_size[0], strides[0],
		padding), filters)
		except:
		pass

		@@ -1993,13 +2003,13 @@ class Conv2D(SharedVariableScope):
		strides = stride
		if isinstance(stride, int):
		strides = (stride, stride)
		if padding.lower() == 'same':
		if isinstance(kernel_size, int):
		kernel_size = (kernel_size, kernel_size)
		self._shape = (parent_shape[0],
		int(np.ceil(parent_shape[1] / strides[0])),
		int(np.ceil(parent_shape[2] / strides[1])), num_outputs)
		else:
		self._shape = (parent_shape[0], parent_shape[1] // strides[0],
		parent_shape[2] // strides[1], num_outputs)
		_conv_size(parent_shape[1], kernel_size[0], strides[0],
		padding),
		_conv_size(parent_shape[2], kernel_size[1], strides[1],
		padding), num_outputs)
		except:
		pass

		@@ -2115,15 +2125,15 @@ class Conv3D(SharedVariableScope):
		strides = stride
		if isinstance(stride, int):
		strides = (stride, stride, stride)
		if padding.lower() == 'same':
		if isinstance(kernel_size, int):
		kernel_size = (kernel_size, kernel_size, kernel_size)
		self._shape = (parent_shape[0],
		int(np.ceil(parent_shape[1] / strides[0])),
		int(np.ceil(parent_shape[2] / strides[1])),
		int(np.ceil(parent_shape[3] / strides[2])), num_outputs)
		else:
		self._shape = (parent_shape[0], parent_shape[1] // strides[0],
		parent_shape[2] // strides[1],
		parent_shape[3] // strides[2], num_outputs)
		_conv_size(parent_shape[1], kernel_size[0], strides[0],
		padding),
		_conv_size(parent_shape[2], kernel_size[1], strides[1],
		padding),
		_conv_size(parent_shape[3], kernel_size[2], strides[2],
		padding), num_outputs)
		except:
		pass

Admin message