Open shilpamatne opened 1 year ago
If you want to create the CAM only for the RGB part,
I would create a model wrapper that fixes the D input, and is just a regular model with an RGB input.
Something like
class RGBDWrapper:
def __init__(model, depth_tensor):
self.model = model
self.depth_tensor = depth_tensor
def __call__(rgb_tensor):
self.model(rgb_tensor, depth_tensor)
wrapped_model = RGBDWrapper(depth_tensor)
Then from the point of view of the CAM model, this wrapper model is just a model that accepts an RGB tensor.
Can I use this package with RGBD images? My custom model is a two stream network taking RGB image in one stream and depth image in the other. How can I specify input_tensor in such a case? Kindly advise.