Tensor dimension mismatch in pre-trained model (channel related)
#23
by zuttng - opened
Hello, I have a question. Do the weights in the pre-trained model only support six-channel images? My four-channel image input results in a 1025-dimensional image (4×256+1), but the model tells me the expected dimension is 1537 (6×256+1). Does the pre-trained model support the parameter "channel_agnostic"?
Thank you！