search

jankolf / GraFIQs

Official repository of paper "GraFIQs: Face Image Quality Assessment Using Gradient Magnitudes"
https://arxiv.org/pdf/2404.12203.pdf
15 stars 1 forks source link

Gives empty .txt files #1

Closed ogladkov closed 6 months ago

ogladkov commented 6 months ago

I set the run like that: python extract_grafiqs.py --image-path /home/sm00th/Projects/assets/IMG_3625.jpg --output-dir /home/sm00th/Projects/apps/GraFIQs/output/ --backbone iresnet100 --weights /home/sm00th/Projects/apps/GraFIQs/backbones/resnet100_ms1mv2_arcface.pth

But got empty text files in the output dir...

2024-05-07_14-40

jankolf commented 6 months ago

Hi Oleg, could you please try and run the script with the following command: python extract_grafiqs.py --image-path /home/sm00th/Projects/assets/ --output-dir /home/sm00th/Projects/apps/GraFIQs/output/ --backbone iresnet100 --weights /home/sm00th/Projects/apps/GraFIQs/backbones/resnet100_ms1mv2_arcface.pth --image-extension jpg

The script currently does not read a single file, only an entire folder of images that contains images with the extension --image-extension.

Best, Jan

ogladkov commented 6 months ago

@jankolf thanks! it works! But how to interpret the results? I got: assets/IMG_0594.jpg 0.022407490760087967 assets/IMG_0631.jpg 0.024729223921895027 assets/IMG_0716.jpg 0.018773064017295837 assets/IMG_0807.jpg 0.03962496668100357 assets/IMG_1478.jpg 0.02287452295422554 assets/IMG_1506.jpg 0.029470067471265793 assets/IMG_1613.jpg 0.020342953503131866 assets/IMG_1780.jpg 0.029832670465111732 assets/IMG_1849.jpg 0.03149200230836868 assets/IMG_1986.jpg 0.022768206894397736 assets/IMG_2023.jpg 0.02831069380044937 assets/IMG_2236.jpg 0.02534307725727558 assets/IMG_2265.jpg 0.020838480442762375 assets/IMG_2477.jpg 0.026965424418449402 assets/IMG_2924.jpg 0.021429385989904404 assets/IMG_7829.jpg 0.02555595338344574 assets/IMG_7830.jpg 0.029234034940600395 assets/IMG_7850.jpg 0.025051038712263107 assets/IMG_7851.jpg 0.020996196195483208 assets/IMG_8684.jpg 0.026479508727788925

jankolf commented 6 months ago

The lower the score, the better the image is suited for FR, so the face image was assigned a higher quality. A higher score indicates that the face image was assigned lower quality and is less suited to perform FR.

ogladkov commented 6 months ago

Thank you @jankolf ! Now is clear!