1 paper · 2 filters
Jiahao Wang, Ting Pan, Haoge Deng +4
Autoregressive models with continuous tokens form a promising paradigm for visual generation, especially for text-to-image (T2I) synthesis, but they suffer from high computational…