Are there up-to-date resources that compare different neural networks in terms of "quality"?

Question

Neural networks

Are there resources that compare different neural networks (solving tasks of the same type) with open source code (for example, on github) in terms of speed, performance, and so on, roughly speaking, quality.
That is, approximately, I mean: at the moment, the code from the repositories X1, X2, X3 (with such and such config parameters) is best handled by task A, Y1, Y2, Y3, etc., with task
B networks: on the github there are already quite a large number of varieties of the same GANs (WGAN, BEGAN, etc.), autoencoders (VAE), etc.
And more and more new ones continue to appear like mushrooms after the rain
. The situation on the topic of super resolution is more or less clear - so far, as far as I understand, ESRGAN is leading there.
True, a cursory test of finished models somehow didn’t impress me much, as they did, for example, NVidia pggans in terms of generating 1024x1024 faces.
For example, you can get information from various papers on arxiv.org.
But they do not always compare their algorithm with others or compare with objectively weak ones, and you have to look for information in the entire article or through Google (for example, the request "than SRGAN" sends to ESRGAN), and so on, but this information may not be.
Ideally, of course, I would like to have a finite number of such sources within reasonable limits (for example, up to a maximum of 10), so as not to jump back and forth.
Where can I find similar resources with a comparison (at least approximate) of different networks?

Answer 1

You can compare not on "problems of the same type", but on the solution of a specific problem on a specific data set. What, in fact, kaggle does. And there is no universal answer to this question.
In general, many perceive machine learning as "take a ready-made solution -> apply _> get a super result." In fact, it is always deep research and professional analysis.
As an example, I can give an analysis of the applicability and effectiveness of various forecasting methods in time series (one of the most famous and relevant tasks) . Like - everything is clear. However, try to make sense of the results without much prior study of the theory :-)
https://journals.plos.org/plosone/article/file?id=...

Answer 2

Google "sota" (state-of-the-art)
You can find, for example, this: https://github.com/RedditSota/state-of-the-art-res...