Description
Description
Eg if using a 4 channel dataset, the error will be raised:
File "/usr/local/lib/python3.11/site-packages/torchvision/models/detection/transform.py", line 141, in forward
image = self.normalize(image)
^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.11/site-packages/torchvision/models/detection/transform.py", line 169, in normalize
return (image - mean[:, None, None]) / std[:, None, None]
~~~~~~^~~~~~~~~~~~~~~~~~~~~
RuntimeError: The size of tensor a (4) must match the size of tensor b (3) at non-singleton dimension 0This is because GeneralizedRCNNTransform only supports 3 channels. We want to use Kornia for norm instead or disable this behaviour from torchvision
Steps to reproduce
I'm using a proprietary 4 channel dataset
model:
class_path: ObjectDetectionTask
init_args:
model: faster-rcnn
backbone: resnet18
weights: True
lr: 5e-4
in_channels: 4
Version
torchgeo==0.7.0