Papers
7
Total Citations
424
H-Index
5
About
Junda Cheng is a rising star in computer vision, whose research focuses on a core challenge for robotics and autonomous systems: stereo matching. His work has fundamentally advanced how machines perceive depth from two images, achieving a powerful blend of accuracy and efficiency. Cheng’s most significant contribution is the **Attention Concatenation Volume (ACV)** , a novel cost volume construction method that intelligently weights matching information. This innovation, detailed in his highly-cited 2022 paper (269 citations), set a new standard for the field. He further refined this approach in a 2023 journal article (84 citations), demonstrating its robustness. Building on this foundation, Cheng developed **IGEV++** (2025, 37 citations), an iterative architecture that uses multi-range geometry encoding volumes to resolve matching ambiguities in challenging regions like large disparities and textureless areas. His work **Coatrsnet** (2023) also explores the synergistic use of convolution and attention for stereo matching. With over 420 total citations in just a few years, Junda Cheng is a leading voice in geometric deep learning, providing the foundational algorithms that help machines see the world in 3D.
Research Focus
Key Achievements
Top Papers
- 1Attention Concatenation Volume for Accurate and Efficient Stereo Matching269 citations · 2022
- 2Accurate and Efficient Stereo Matching via Attention Concatenation Volume84 citations · 2023
- 3IGEV++: Iterative Multi-Range Geometry Encoding Volumes for Stereo Matching37 citations · 2025
- 4
- 5Attention Concatenation Volume for Accurate and Efficient Stereo Matching10 citations · 2022
- 6
- 7