search

jinnh / GSAD

[NeurIPS 2023] Global Structure-Aware Diffusion Process for Low-Light Image Enhancement
121 stars 9 forks source link

NIQE codes #6

Closed Xingshi-tech closed 11 months ago

Xingshi-tech commented 11 months ago

I found some methods online for calculating the NIQE metric, and I used them to test the enhancement results, but the results I got were inconsistent with the ones provided in your paper. If possible, could you please share the code you used to calculate the NIQE?

ZHU-Zhiyu commented 11 months ago

Hi @Xingshi-tech , Thanks for your interest. We will release the corresponding code soon (before this weekend).

Xingshi-tech commented 11 months ago

Thanks for reply

jinnh commented 11 months ago

Hi, @Xingshi-tech. For NIQE codes, please refer to metrics in Lighting-the-Darkness-in-the-Deep-Learning-Era-Open.

When testing the unpaired data (e.g., LIME, NPE, MEF, DICM, and VV), you can try to use different pre-trained models (It is recommended to use 'lolv2_syn_gen.pth') and employ various sampling steps and noise levels to obtain visual-pleasing results by modifying these terms in the 'test_unpaired.json'.

For example, when testing LIME, we use 'lolv2_syn_gen.pth' with the sampling setting "val": { "schedule": "linear", "n_timestep": 10, "linear_start": 1e-4, "linear_end": 7e-1 }.

Xingshi-tech commented 10 months ago

Hi,@jinnh. Thanks for your help.