A plug-and-play pytorch module to implement custom filtering for images.
Sometimes, we need to implement some image filtering operations in the neural network. Therefore, we design this simple torch_filter module.

filter_weight: the custom filter weights. Note that both dimensions of the filter must be equal and odd. eg. (3×3), (5×5). Type must be np.array.
is_grad: True or False. Whether this module is involved in backpropagation.

The test.py implements image sharpening operation.

filter_weight=np.array([[-1, -1, -1], [-1, 9, -1], [-1, -1, -1]])