facebookresearch / DiT

Official PyTorch Implementation of "Scalable Diffusion Models with Transformers"
Other
6.37k stars 569 forks source link

Question on Evaluation #94

Open Ting-Justin-Jiang opened 4 months ago

Ting-Justin-Jiang commented 4 months ago

Thank you so much for the amazing work. When calculating FID with an off-the-shelf XL/2 512 model, I generated 10K images (cfg=1.5) and obtained the following results:

Inception Score: 243.23751831054688 FID: 6.0483700680604215 sFID: 18.34898058428621 Precision: 0.8374 Recall: 0.621

which doesn't align with any benchmark result provided. Is it because I sampled too few images?