convert_from_tensorflow.py: add support when kernel size is 1*1 with one input/output channel (gray image)

Guo, Yejun committed 6 years ago

Commit e52070e1 parent ed9fc2e1 file · +4 −1Browse files

Unified · Split

tools/python/convert_from_tensorflow.py+4−1

@@ -193,7 +193,10 @@ class TFConverter: · 192 unmodified lines
193	193	filter_width = ktensor.tensor_shape.dim[1].size
194	194	in_channels = ktensor.tensor_shape.dim[2].size
195	195	out_channels = ktensor.tensor_shape.dim[3].size
196		− kernel = np.frombuffer(ktensor.tensor_content, dtype=np.float32)
	196	+ if filter_height * filter_width * in_channels * out_channels == 1:
	197	+ kernel = np.float32(ktensor.float_val[0])
	198	+ else:
	199	+ kernel = np.frombuffer(ktensor.tensor_content, dtype=np.float32)
197	200	kernel = kernel.reshape(filter_height, filter_width, in_channels, out_channels)
198	201	kernel = np.transpose(kernel, [3, 0, 1, 2])
199	202