essunny310 / FreestyleNet

[CVPR 2023 Highlight] Freestyle Layout-to-Image Synthesis
https://essunny310.github.io/FreestyleNet/
MIT License
149 stars 3 forks source link

Results of Evaluation #14

Open linyq2117 opened 7 months ago

linyq2117 commented 7 months ago

Hi, thanks for your great work!

I try to generate images using your provided checkpoints and run sample_ADE20K.sh. However, the generated images seem a little different from your provided one here. I got 26.2 FID and 37.0 mIoU, which are different from those reported in the paper.

If i understand right, the results should be consistent under the fixed checkpoints and seeds. Is the provided model the best one used in your paper, or did you use a different setting during image generation (e.g., random seed, batch size, sampler)?