🏢 Fudan University

VidCRAFT3 enables high-quality image-to-video generation with precise control over camera movement, object motion, and lighting, pushing the boundaries of visual content creation.

VideoRoPE enhances video processing in Transformer models by introducing a novel 3D rotary position embedding that preserves spatio-temporal relationships, resulting in superior performance across var…

Agent-R: A novel self-training framework enables language model agents to learn from errors by dynamically constructing training data that corrects erroneous actions, resulting in significantly improv…

DOLPHIN: AI automates scientific research from idea generation to experimental validation.

LiFT leverages human feedback, including reasoning, to effectively align text-to-video models with human preferences, significantly improving video quality.

BitStack: Dynamic LLM sizing for variable memory!