Had an issue when integrating perceiver architechture that all the implementations require channels to be the last input dim of the minibatch. This required me transposing after OpToTensor in isic dataloader.
This can also be done using OpLambda and passing torch.transpose just as easily, so its not a very necessary pull request. Only merge this if you want to @mosheraboh
Had an issue when integrating perceiver architechture that all the implementations require channels to be the last input dim of the minibatch. This required me transposing after OpToTensor in isic dataloader.
This can also be done using OpLambda and passing torch.transpose just as easily, so its not a very necessary pull request. Only merge this if you want to @mosheraboh