InstructIR

Why we included this project

Photographers and computer vision engineers often end up with images that are noisy, dark, or blurred, and fixing each problem usually means switching tools. InstructIR takes a different route: you give it a photo and a plain-language instruction, and a single network handles denoising, deraining, deblurring, dehazing, and low-light enhancement. That makes it handy for teams that want one pipeline for many cleanup jobs, and the demo lets you try requests like tone mapping or a more professional look. The repo includes pretrained weights, evaluation scripts, and a Colab notebook, so you can reproduce the reported results or plug the model into your own workflow.