I am new to computer vision research (e.g. image classification) and I want to implement some ideas to a model to see whether it increases the performance. However there are too many models on the paperwithcode, should I just pick the top-performance one or carefully choose one? I saw there are some people complaining about unconvergence and overfitting problems in some of the implementation on the Github. Is there a suitable (stable and easy-to-use) model for CV research beginners to implement and experiment on? Oh, by the way, I am using PyTorch.
Thanks.