- Research Roadmap Slides [中文版] [English version]
- Machine Perception of Human Activities
- Efficient long-term action detection in extended videos [CMU-DIVA]
- End-to-end action detection [Stargazer]
- Multi-modal multi-dataset model co-training [NeurIPS'22, NeurIPS'22]
- Zero-shot / Few-shot action recognition
- First-person / ego-centric view action recognition & 3D action recognition & Viewpoint invariant representation [Example]
- Learning from 3D simulation [ForkingPaths], [SimAug], [CARLA Sim]
- Future Prediction
-
Action Anticipation & Human Intention Prediction [Next-prediction]
- Sub-direction: Predictive Self-supervised Learning
- Sub-direction: Video Generation, Generative Model - Diffusion Model
- Trajectory Prediction [Multiverse]
- Time-series Forecasting (weather, energy, economics, etc.)
-
Action Anticipation & Human Intention Prediction [Next-prediction]
- AI + X
- Aerial Video analysis [natural disaster assessment using drone videos - WACV’21]
- Robotic Helper (Robotic Third Hand) [NSF grant]
- Medical Image Analysis [3D semantic segmentation with cryo-ET images (for protein molecule)]
- Edge Computing
- Efficient PPL/Model Design [ODT]
- Knowledge Distillation
Figure 1. Project overview of the Precognition group (09/2022)
