1 paper · 1 filter
Hengyu Zeng, Xin Gao, Guanghao Li +5
Continuous image tokenizers enable efficient visual generation, and those based on variational frameworks can learn smooth, structured latent representations through KL regularizat…