Make the output_mapper support loading images from List of PIL image, or List of string.
Why ?
When you try to use torchvision datasets, the output of the dataset is PIL images. If the user passes a Dataloader, the output will be a list of PIL images, which is currently not supported. The same issue could happen if the dataset outputs a string, then we would not support the Dataloader.
Extra
I also added some extra fixes to make the unit tests properly work with the CI.
What ?
Make the output_mapper support loading images from List of PIL image, or List of string.
Why ?
When you try to use torchvision datasets, the output of the dataset is PIL images. If the user passes a Dataloader, the output will be a list of PIL images, which is currently not supported. The same issue could happen if the dataset outputs a string, then we would not support the Dataloader.
Extra
I also added some extra fixes to make the unit tests properly work with the CI.