Yes, there are plenty of them. Not sure what the SOTA is, though. Similar to super-resolution, it is relatively simple to create a nearly-infinite dataset for these; pick a clean image, then introduce JPG artifacts. Then train a model to invert the process.