about
Interested Areas
High Performance Computing, Mobile Computing, Computational Photography, Low-Level Computer Vision, Mobile Development, Deep Learning, GPU Programming
ByteDance(TikTok), (2023-12 to Present)
Software Engineer
San Jose, CA
• Developed TikTok’s in-app camera stabilization feature and launched global A/B testing to select the best strategy that maximized user publishing gains.
• Implemented in-app camera HDR capture based on the OpenGL ES rendering pipeline, handling HDR to SDR conversion and HDR color space transformation.
• Upgraded TikTok’s in-app camera to support High Resolution image capture, replacing previewed frame screenshots with the system’s native camera API for improved image quality.
• Optimized camera start latency.
SenseBrain, (2022-03 to 2023-11)
Senior Software Engineer
San Jose, CA
• Designed and developed high performance mobile imaging pipeline on Qualcomm and MediaTek platform.
• Optimized algorithms for mobile CPUs using C++ and ARM Neon intrinsics.
• Optimized algorithms for mobile GPUs using OpenCL.
• On-device deep neural networks deployment with neural inference engines such as SNPE, QNN, PPL and libTorch.
• Compressed Generative AI model (stable diffusion) for mobile devices by implementing quantization and deployed it on Android devices featuring the Qualcomm 8550 SoC using QNN, resulting in a runtime of 12 seconds.
• Developed customized application for camera sensor data acquisition and processing, enabling features like burst raw image capture with 3A control, multi-camera streaming and dual-camera fusion on the Android platform.
SenseBrain, (2021-11 to 2022-03)
Software Engineer
San Jose, CA
• Managed the maintenance and enhancement of NightSight and rawHDR imaging pipelines for mobile platforms
• Enhanced algorithm efficiency and performance on mobile GPUs and CPUs using OpenCL and SIMD techniques.
• Developed learning-based image denoising methods based on UNet and SGN architecture under low-light condition.
• Improved denoising performance by applying controllable priors such as segmentation maps and variance maps during
training and inference.
• Camera sensor noise modeling, calibration and synthesis under low-light condition.
• Developed multi-frame noise reduction and multi-scale DCT denoising algorithm for Bayer and RGBW sensor.
SenseBrain, (2021-05 to 2021-11)
Research Intern
San Jose, CA
• Developed and experimented lightweight deep networks for raw domain image denoising.
• Synthetic noise modeling and calibration under low-light condition.
Publication
google scholar
Purdue University
2018 - 2020
Master of Science
Computer Graphics
Advisor: Yingjie Chen
