image VS rust-ndarray

It uses https://github.com/1Password/arboard and https://github.com/image-rs/image under the hood. I haven't tested deeply but I assume it should handle many formats out of the box.

You can use the image crate.

As for image processing, I'd start on crates.io and search for "image processing" and skim through to find the most downloaded crates -- image seems to be the most popular, with over 7 million downloads, so I'd check if that has what you want, and if not photon-rs seemed relatively mature with perhaps more features (albeit far less commonly used)

Glad it worked out for you—it seemed like the features are there for basic image conversion, but try to do something a little more sophisticated and you run into some severe design limits. Here is the kind of underlying problem I’m talking about:

https://github.com/image-rs/image/issues/1523

Side note the image-rs crate in the Rust ecosystem has no external dependencies and can encode and decode animated gifs and a variety of other image formats, 100% rust https://github.com/image-rs/image

I believe that photon uses image internally which only provides the ability to overlay another image. The crate imageproc extends image and adds the draw_text method.

And for image resizing I used the "image" crate: https://crates.io/crates/image. Note that while in development mode resizing an image takes a long time (several seconds for me), but in production mode it is lightning fast. I spent some time trying to figure this out :D

Why not just add support for QOI to image?

The image create is actually pretty good. It decodes a lot of stuff natively, and encodes to JPEG and PNG. I actually used it to make something very similar.

the image crate might be just what you're looking for. more specifically, the gif encoder can do exactly what you want.

Burn is different: it is built around the Backend trait which encapsulates tensor primitives. Even the reverse mode automatic differentiation is just a backend that wraps another one using the decorator pattern. The goal is to make it very easy to create optimized backends and support different devices and use cases. For now, there are only 3 backends: NdArray (https://github.com/rust-ndarray/ndarray) for a pure rust solution, Tch (https://github.com/LaurentMazare/tch-rs) for an easy access to CUDA and cuDNN optimized operations and the ADBackendDecorator making any backend differentiable. I am now refactoring the internal backend API to make it as easy as possible to plug in new ones.

Looks like it's an open request

https://github.com/rust-ndarray/ndarray/issues/281

Answer: you can’t with this crate. I implemented a dynamic n-dim solution myself but it uses views of integer indices that get copied to a new array, which have indexes to another flattened array in order to avoid duplication of possibly massive amounts of n-dimensional data; using the crate alone, copying all the array data would be unavoidable.

Ultimately I’ve had to make my own axis shifting and windowing mechanisms. But the crate is still a useful lib and continuing effort.

While I don’t mind getting into the weeds, these kinds of side efforts can really impact context focus so it’s just something to be aware of.

In principle you're trying to convert between columnar and row-based data layouts, something that happens fairly often in data science. I bet there's some hyper-efficient SIMD magic that could be invoked for these slicing operations (and maybe the iterator solution does exactly that). Might be worth taking a look at how the relevant Rust libraries like ndarray do it.

ndarray and numpy crates provide good way to operate on numpy ndarrays from python

I don't think any of the major ML projects have GPU acceleration because ndarray doesn't support it.

Not sure about ndarray: https://github.com/rust-ndarray/ndarray/issues/840

I used basic_dsp a while back and found it lacking. I was hoping to find something that uses the ndarray datatype but i'm not seeing this yet. If you're primarily trying to learn though it might not really matter which library you contribute to. As for myself, I just picked the one that was most used and actively worked on at the time. However I keep an eye out on other libraries; if I see something take off, I might switch over. Either way you'll learn and can point to it as work accomplished.

Most libraries use either an older C/C++ dynamically or statically. I was looking into maybe rewriting some image manipulation functions in Rust using ndarray (which would then allow for both hardware, gpu and multithreading acceleration), but it would take a very long to get fully into the subject.

The best way would to use a crate for this, such as ndarray, which provides some good ways to do this.