Projects & Achievements
Learnable Image ISP Pipeline
Image ISP is a critical processing unit that maps RAW images from the camera sensor to RGB images. The traditional ISP pipeline requires manually tuning each sub-processing unit that takes legions of efforts from imagery experts. To tackle with this downside, we design a novel network to enable a learnable ISP pipeline that enhances smartphone images and mitigates artifacts from the camera sensor. In AIM2020 learned image ISP challenge, our proposed solution outperformed the state-of-the-arts and achieve very high MOS (4.5 out of 5) while remaining competitive in numerical results (21.86dB in PSNR). See more details [here].
Single & Burst Image Demoiring
We proposed a new network, named Cube-DemoireNet, which enhances the conventional multi-scale network by 1) facilitating information exchange across different scales and stages to alleviate the underlying bottleneck issue, and 2) employing the attention mechanism to identify the dynamic Moiré patterns, and thus eliminate the ones effectively with the preservation of image texture. See more details [here].
Snake Game with Deep Q-Learning
A snake game with an AI agent. This program is designed in a way that both users or AI can take in control so that enables users to compete with AI. This project is mainly built on PyTorch and PyGame. See more details [here].
- Runner-Up Award in AIM 2020 Challenge on Learned ISP Track 2: Perceptual [Certificate]
- NTIRE 2020 Challenge on Image Demoireing Burst Track - 3rd Place
- Vector Institude Scholarship in Artificial Intelligence (VSAI) 2019
- Provost’s Honour Roll 2019 (GPA 12 out of 12)
- Dean’s honour list 2016-2019