多模态理解算法实习生

- Analyze multimodal model capability limits - Build high quality training data for foundation models - Build video understanding datasets for human behavior event recognition video caption temporal grounding - Construct multimodal data pipelines - Develop embodied understanding models for physical world - Improve model perception and cognition in visual localization - Label chain of thought reasoning and build evaluation sets - Support long horizon behavior and task decomposition - Synthesize clean annotate format and evaluate text image video and action data