5 papers
UniCom: Unified Multimodal Modeling via Compressed Continuous Semantic Representations
Yaqi Zhao, Wang Lin, Zijian Zhang +5
Current unified multimodal models typically rely on discrete visual tokenizers to bridge the modality gap. However, discretization inevitably discards fine-grained semantic informa…
Demystifying Numerosity in Diffusion Models -- Limitations and Remedies
Yaqi Zhao, Xiaochen Wang, Li Dong +2
Numerosity remains a challenge for state-of-the-art text-to-image generation models like FLUX and GPT-4o, which often fail to accurately follow counting instructions in text prompt…
SEA: Supervised Embedding Alignment for Token-Level Visual-Textual Integration in MLLMs
Yuanyang Yin, Yaqi Zhao, Yajie Zhang +7
Multimodal Large Language Models (MLLMs) have demonstrated remarkable capabilities by integrating visual and textual inputs, yet modality alignment remains one of the most challeng…
Towards Precise Scaling Laws for Video Diffusion Transformers
Yuanyang Yin, Yaqi Zhao, Mingwu Zheng +11
Achieving optimal performance of video diffusion transformers within given data and compute budget is crucial due to their high training costs. This necessitates precisely determin…
Beyond Sight: Towards Cognitive Alignment in LVLM via Enriched Visual Knowledge
Yaqi Zhao, Yuanyang Yin, Lin Li +7
Does seeing always mean knowing? Large Vision-Language Models (LVLMs) integrate separately pre-trained vision and language components, often using CLIP-ViT as vision backbone. Howe…