top of page



I work on efficient AI perception solutions with Dense Supervision approaches :

1. 3D Perception on embedded platforms: My research team at RI CMU empowers cameras and sensors on moving platforms-autonomous machines, drones, smart camera modules-with (a) real-time 3D reconstruction capability for persons, objects, and scenes (3) the light-weight Deep Neural Networks (DNNs) that consume less training data, computation, and deployment efforts than standard deep learning approaches. We are making intelligent perception solutions accessible to general industries and markets.

2. Perception in multi-modality systems: We develop deep learning approaches that take multiple sensor data (e.g., videos, depth, wearable sensor signal, and WiFi signal) to produce multiple perception results, such as location, poses, activities, and physical metrics. Our deep approaches also trace the key factors of perception results for the purpose of improving the psychical performance of athletes and quality-of-life for the general public.


Google Scholar Page
  • LinkedIn

Current Sponsors/Partners:



Teaser videos


Search video...
Play Video


Play Video



May 2024: "Mono WiFi Scene" and "Hello 3D" to appear at ICRA2024 (video link pending). 
May 2024: "Hello 3D" demo on the Automate 2024 trade show Linkedin Post
Oct. 2023: Introducing "Hello 3D", a universal software engine for real-time 3D human perception in moving robots, stationary monitoring, and sports training, Project link
Sep. 2023: Introducing "Mono WiFi Scene" the first technology that makes it possible in dense sensing of 3D environment using only a single(mono) WiFi device, Project link.
Jun. 2023: Our work on "Dense Pose from Wi-Fi" is recently covered by the Economist, BBC News Digital Planet, New Scientist, and Synced.  
Aug.2021: Tech news on our WiFi sensing technology on Post-Gazette News Link
Jul. 2021: Elaborative Rehearsal for Zero-shot Action Recognition accepted by ICCV 2021!
Mar. 2021: Optimal Gradient Checkpoint Search for Arbitrary Computation Graphs accepted by CVPR 2021 (Oral)! 

Sep. 2020: Comprehensive Attention Self-Distillation (CASD) accepted by NeurIPS 2020!
Aug. 2020: RSC and MAL codes released in our [Github]
Jun. 2020: Representation Self-Challenging (RSC) accepted by ECCV 2020 Oral (top 2%)! 
Feb. 2020: Multiple Anchor Learning (MAL) accepted by CVPR 2020! 
Sep. 2019: Inverted Attention (IA) accepted by WACV 2020! 
Jul. 2019: Person-in-WiFi accepted by ICCV 2019! 
May. 2019: 9 Papers on Deep Learning Research 
Mar. 2019: DeLight Group at Robotics Institute, CMU
bottom of page