The current model has mostly been tested on high-range float32 images, so there may be other images or formats where the model is not able to align the input images perfectly.

The simplest way to contribute to this project is by finding images that cannot be properly aligned with the align() or fast_align() functions. If that happens, please do not immediately flee to the next repository and take the time to let me know.

The code could be optimized to perform the same tasks using fewer lines of code.

Any ideas & suggestions are always welcomed.