Export
Yunchao Ma
· 3 previous affiliations
- Alibaba Group (China)Since 2022
- Researcher · Since 2022
- Megvii (China)2018–2022
- Researcher · 2018–2022
- Harbin Institute of Technology2016–2018
- MS student · 2016–2018
Research Overview
Yunchao Ma focuses on computer vision and computational photography, specifically addressing depth estimation and completion tasks. The available evidence indicates a concentrated effort in 2023 on improving depth map resolution and structural accuracy using deep learning techniques. Recent work involves refining monocular depth estimation models and participating in community-wide evaluations of RGB and Time-of-Flight sensor fusion.
Research Trajectory
The available corpus of work for Yunchao Ma is concentrated in 2023, focusing primarily on depth estimation and completion within computer vision. A central problem addressed in this period is the refinement of depth maps generated by deep learning models, which often struggle with preserving fine details and maintaining correct global depth layouts when applied to high-resolution images.
As a second author, Ma contributed significantly to algorithmic developments in this area with Promoting Monocular Depth Estimation by Multi-Scale Residual Laplacian Pyramid Fusion. This research introduced a multi-scale residual Laplacian pyramid fusion network (MS-RLap-FNet) designed to fuse multi-scale depth maps from existing estimation models. The approach aimed to overcome the limitations of standard deep networks by generating high-quality depth maps with richer details and improved structural integrity.
In addition to targeted methodological improvements for monocular depth estimation, Ma has participated in broader community evaluations concerning multi-modal sensor fusion. As a contributing researcher among a large group of authors, Ma was involved in the MIPI 2023 Challenge on RGB+ToF Depth Completion: Methods and Results. This collaborative effort benchmarked methods for completing depth maps using a combination of standard RGB images and sparse Time-of-Flight (ToF) measurements. The challenge provided a standardized dataset and evaluation metrics to assess the accuracy and efficiency of emerging deep learning approaches, reflecting an active engagement with the practical challenges of deploying depth completion systems in robotics and advanced computer vision applications.
Profile updated
| Title | Venue | Year |
|---|---|---|
| Promoting Monocular Depth Estimation by Multi-Scale Residual Laplacian Pyramid Fusion | IEEE Signal Processing Letters | 2023 |
| MIPI 2023 Challenge on RGB+ToF Depth Completion: Methods and Results | Workshops @ CVPR | 2023 |
Loading new papers…
Loading research neighbors…
See something wrong? Report a mistake