search

ZhexinLiang / CLIP-LIT

[ICCV 2023, Oral] Iterative Prompt Learning for Unsupervised Backlit Image Enhancement
https://zhexinliang.github.io/CLIP_LIT_page/
269 stars 23 forks source link

How to modify if training other tasks #12

Open Zachary-ZCH opened 10 months ago

Zachary-ZCH commented 10 months ago

I want to try the performance on other image enhancement tasks. Can you give me some suggestions on what needs to be modified?

ZhexinLiang commented 10 months ago

Hi @Zachary-ZCH, thanks for your interest in our work. Sorry for my late response. I am currently a little bit busy on my next project.

I think you should modify the parameters and the network. You can use a network commonly-used in the tasks you want to try. And for the parameters, based on my experiments, num_reconstruction_iters, num_clip_pretrained_iters, the train_thre and the total_thre are very important for the training.

You can also modify this parameter and the parameters in the margin loss if you want.

If you have any other questions, feel free to discuss with me.