search

gligen / GLIGEN

Open-Set Grounded Text-to-Image Generation
MIT License
1.98k stars 148 forks source link

Evaluation code #28

Open CaoZelin12138 opened 1 year ago

CaoZelin12138 commented 1 year ago

The evaluation code for calculating FID and YOLO-score(AP) is not provided. Can you publish the code on github?

Yuheng-Li commented 1 year ago

For FID, we are using the https://github.com/mseitzer/pytorch-fid For YOLO-score, we are using https://github.com/Tianxiaomo/pytorch-YOLOv4 (Note: there is a typo in their code; please check this issue before running their code)

qqphung commented 1 year ago

How to calculate AP with generated images, which don't have objects in the groundtruth. For example: groundtruth bounding box: dog, but the generated image: cat

qqphung commented 1 year ago

Can @CaoZelin12138 reproduce the FID score in COCO? I sample 30k captions and get ~15, which is higher than paper report.

DiamondsH commented 7 months ago

How to calculate AP with generated images, which don't have objects in the groundtruth. For example: groundtruth bounding box: dog, but the generated image: cat

hello？have you resolved this problem? i don't find the ground truth of the boxes ,either! Wish your reply!thanks!