multimodal-modeling

该论文挑战了离散视觉令牌必然存在语义损失的假设，提出因子化向量量化（Factorized Vector Quantization），在多模态理解上达到与连续视觉表示相当的水平，并在离散表示模型中取得最先进性能。

UniSpace introduces a unified visual representation that unifies semantic understanding, high-fidelity reconstruction, and image generation in a single space using a reparameterized ViT, eliminating the need for a separate VAE.