I dabbled in the research of computer vision, a branch of CS, before. Most of the papers published do not have any codes attached, and given the "black magicnes

I dabbled in the research of computer vision, a branch of CS, before. Most of the papers published do not have any codes attached, and given the "black magicness" of many of the training approaches, a lot of the papers are hard to replicate. In fact, I find it dubious that many papers claim significant improvement when the same amount of improvement can be brought about by simply better parameter tuning. Without codes I can never be sure that their methods actually worked.