The shape of conv_layer_output_value is (1, 14, 14, 512), and the shape of pooled_grads_value is (512, ).
So it will cause IndexError when multiple conv_layer_output_value[:, :, i] with pooled_grads_value[i] in the source code.
Thus I change it to conv_layer_output_value[:, :, :, i] to align the dimension.
The shape of
conv_layer_output_value
is (1, 14, 14, 512), and the shape ofpooled_grads_value
is (512, ).So it will cause IndexError when multiple
conv_layer_output_value[:, :, i]
withpooled_grads_value[i]
in the source code. Thus I change it toconv_layer_output_value[:, :, :, i]
to align the dimension.