Happens to me too, even when passing n_in=1 to the learner
I have to say that I eventually gave up on debugging, and processed the data to have the appropriate number of channels (by multiplying the single channel). It’s a crappy solution, and I’m still working on improving it, but it worked.
Is there a plan to have ImageImageDataLoaders in v2? Or is there a simple way to achieve image to image data loading currently? I also would like to have a network with images as input and output.
Good job! Was this to replace the whole RandomErasing function or something in it?
It was for a Denoising Autoencoder experiment: fastai2_projects/Denoising AutoEncoder.ipynb at master · jerbly/fastai2_projects · GitHub
1 Like
I see. Nice job.
But was RandNoisyTransform() to replace what we know as RandomErasing() tranform that there is now in the library?