dehazeformer-tiny model for multi-weather image restoration?

[ChangGiMoon]

Thank you very much for your good research. I am a beginner in Image Restoration Task and I have the following questions.

I want to use dehazeformer-tiny model for restoration of various types of weather degradation at near real-time processing speed (using Full-HD resolution). I know that it is desirable to train each type of weather separately and then use the corresponding weights according to the weather.

From my tests, I could not find a faster and more accurate model than dehazeformer-tiny model for haze.

Is it possible to use dehazeformer(tiny model) for other types of weather including haze (e.g. rain, snow)? Specifically, I plan to train by merging different types of weather datasets and then use dehazeformer as a unified model to apply one weight to multiple types of weather changes simultaneously.

Of course, I do not want SOTA performance, and I wonder if this approach is reasonable.

If there is enough room for experimentation, would it be better to sample equally when merging different types of weather? Or would it be better to ignore this and use a lot of samples?

Thank you for reading, and I hope you continue to do good research.