Quantitative evaluations - Githubissues

balthazarneveu / blind-deblurring-from-synthetic-data

MVA ENS Paris Saclay - Image restoration project on deblurring learnt on deadleaves

3 stars 0 forks source link

Quantitative evaluations #9

Open balthazarneveu opened 6 months ago

balthazarneveu commented 6 months ago

At training time, a mix of a bit of everything... just to monitor

1/ Additional metrics

[x] PSNR
[x] SSIM
[x] Perceptual LPIPS Optional "PieAPP" , "NIQE".

2/ Add a test set of natural images

CBSD68 ?
Kodak24 ?
Div2k ?

3/ Add a postprocessing evaluation script: Compute performances depending on:

[x] initiate evaluation script
[x] Depending on noise level

4/ Live metric in interactive pipeline

[x] metrics in natural image inference
[x] metrics in deadleaves script

-> finish with images stored on disk #26