Key Responsibilities

• Design and implement approaches for temporal video understanding and action segmentation in long, untrimmed procedural videos, including:
  • Multi-view action segmentation 
  • Cross-view knowledge transfer 
  • Learning view-invariant representations 
  • Multimodal alignment (e.g., human/hand pose with egocentric and exocentric video streams) 
  • Low-latency models for real-time/online inference 
• Evaluate model performance through rigorous benchmarking and error analysis, and build tools for efficient experimentation, visualization, and reproducibility. 
• Conduct literature reviews, formulate hypotheses, design experiments, and analyze results. 
• Build well-structured, efficient, and reproducible codebases using modern deep learning frameworks (e.g., PyTorch). 
• Collaborate with the team to iterate ideas, models, and system designs. 
• Optionally contribute to research publications (e.g., CVPR, ICCV, ECCV, NeurIPS).  

Minimum Qualifications

• Currently enrolled in an MS or PhD program in Computer Vision, Machine Learning, Robotics, Artificial Intelligence, or a closely related field. 
• Experience with video understanding, including long-form video or temporal modeling, and familiarity with video encoders (e.g., VideoMAE, LaViLa, TimeSformer) and/or fine-tuning approaches. 
• Strong programming skills, with the ability to write clean, efficient, and reproducible code, and proficiency in deep learning frameworks (e.g., PyTorch). 
• Familiarity with model efficiency considerations (e.g., latency, memory, or throughput). 
• Solid problem-solving skills and the ability to independently drive projects from ideation to experimentation (and optionally publication). 
• Strong written and verbal communication skills. 

Bonus Qualifications​