Research
SpecSlot paper proposes pixel-space Transformer with spectral slots for image generation
SpecSlot decouples output detail from global attention length by giving only the lowest-frequency slot per anchor global attention. Reported results include a 2.84 FID gain over a FLOPs-matched JiT control on ImageNet-512 at 3.9% fewer FLOPs.
Every outlet that ran it 1 article
Hugging Face PapersResearchLead
Read, Allocate, Route: Role-Coupled Tokenization for Pixel-Space Generative Transformers