Unified Continuous Generative Models

Open paper
Benchmark Model Rank Results
image-generation-on-imagenet-256x256SiT-XL/2 + UCGM-S (E2E-VAE + 40 sampling steps + CFG)#3FID: 1.06NFE: 80
image-generation-on-imagenet-256x256UCGM-XL/2 (VA-VAE + 30 sampling steps, without guidance)#11FID: 1.21NFE: 30
image-generation-on-imagenet-256x256UCGM-XL/2 (E2E-VAE + 40 sampling steps, without guidance)#12FID: 1.21NFE: 40
image-generation-on-imagenet-256x256LightningDiT + UCGM-S (VA-VAE + 50 sampling steps + CFG)#13FID: 1.21NFE: 100
image-generation-on-imagenet-512x512DDT-XL/2 + UCGM-S (SD-VAE + 150 sampling steps + CFG)#4FID: 1.24NFE: 300
image-generation-on-imagenet-512x512DDT-XL/2 + UCGM-S (SD-VAE + 100 sampling steps + CFG)#6FID: 1.25NFE: 200

Rank counts only results with a code link.