Representation Alignment for Generation: Training Diffusion Transformers Is Easier Than You Think
Open paper
Benchmark
Model
Rank
Results
image-generation-on-imagenet-256x256
SiT-XL/2 + REPA (with the guidance interval)
#29
FID: 1.42
image-generation-on-imagenet-512x512
SiT-XL/2 + REPA (200 epochs, w=1.35) (extra data)
#40
FID: 2.08
Inception score: 274.6
Rank counts only results with a code link.