vision

ConvNeXT's architecture contains an AdaptiveAvgPool2d layer: https://github.com/pytorch/vision/blob/5f03dc524bdb7529bb4f2... This means that you can split your image into tiles, process each tile individually, average the results, apply a final classification layer to the average and get exactly the same result. For reference, see the demonstration below. You could of course do exactly the same thing with a vision transformer instead of a convolutional neural network. That being said, architecture is wildly overemphasized in my opinion. Data is everything. import torch, torchvision.models device = torch.device("cuda" if torch.cuda.is_available() else "cpu") model = torchvision.models.convnext_small() model.to(device) tile_size, image_size = 32, 224 # note that 32 divides 224 evenly image = torch.randn((1, 3, image

Another somewhat related "issue" is that the mean and std used in ImageNet has unknown origins. It's based off the data but it's not clear exactly how they ended up with those numbers. [0] [0] https://github.com/pytorch/vision/issues/1439

haha, yes, but that requires you to modify existing code to do so (which isn't always possible!). There might also be other things you want to do (like add profiling after each op) that would be tedious to do manually, but can easily automated with FX (https://pytorch.org/tutorials/intermediate/fx_profiling_tuto...). Another example is the recent support from torchvision for extracting intermediate feature activations (https://github.com/pytorch/vision/releases/tag/v0.11.0). Like, sure, it was probably possible to refactor all of their code to enable users to specify extracting an intermediate feature, but it's much cleaner to do with FX.

Not quite - it's more of a platform for users to write their own transformations on their PyTorch code. This can include things like doing operator fusion/lowering to a backend compiler, but can also include things like inserting profiling instrumentation (https://pytorch.org/tutorials/intermediate/fx_profiling_tuto...) or extracting intermediate features (https://github.com/pytorch/vision/releases/tag/v0.11.0). Basically, if you want a graph representation of a PyTorch module that's really easy to modify, use torch.fx :)

I was looking for a Python package to manage science datasets, mainly description, download, extraction, on-disk handling, etc... ML frameworks tend to roll their own: - https://github.com/nilearn/nilearn/blob/master/nilearn/datasets/utils.py - https://github.com/tensorflow/datasets/blob/master/tensorflow_datasets/core/downlo… - https://github.com/scikit-learn/scikit-learn/blob/master/sklearn/datasets/base.py - https://github.com/pytorch/vision/blob/master/torchvision/datasets/mnist.py I was hoping that somebody had made a generic package for this purpose so that I don't roll yet another one.

torch.nn offers a very similar interface to Keras (e.g. see Alexnet definition at https://github.com/pytorch/vision/blob/master/torchvision/mo...).