Zechen Bai

A new method to create facial animations efficiently for customized characters using deep learning.

A concise look at hallucinations in MLLMs and strategies to improve reliability.

New methods improve multimodal models by adding questioning and evaluation tasks.

This method adjusts object representation slots based on image complexity.

VideoLISA uses language to segment and track objects in videos effectively.

Factorized Quantization improves image generation through efficient token management.