search

Ephemeral182 / ECCV24_T3-DiffWeather

[ECCV‘24] Teaching Tailored to Talent: Adverse Weather Restoration via Prompt Pool and Depth-Anything Constraint
https://ephemeral182.github.io/T3-DiffWeather
22 stars 0 forks source link

How to add more degredations #1

Open Lincoln20030413 opened 3 days ago

Lincoln20030413 commented 3 days ago

Thank you for your interesting work on the project! I have a question regarding the adaptation of our model to handle additional degradations, such as low-light conditions, blur, and others. Would it be necessary to modify the prompt pool for this purpose? I’d appreciate your insights on the best approach to take.

Ephemeral182 commented 3 days ago

Thanks for your attention! In my view, for more degradations:

  1. You may enlarge prompt pool size to adapt additional degradtions.
  2. We focus on adverse weather restoration, which includes some similarities and difference across different degradations. Thus, such as low-light and noise situation, I consider it will achieve superior performance.
  3. The core motivation of this operation is to utilize unique characteristics and commonalities between degradations. So theoretically, whatever degradations there are, the workflow should be worked. This is because many degradations in the real world often have these properties.
Lincoln20030413 commented 3 days ago

Got it! Thanks