Hengyu Liu

Academic CV

Efficient ML Systems · Visual Computing

Hengyu Liu

Peer-reviewed Publications

  1. Yichi Zhang, Hengyu Liu, Dandan Ding. “A cross-scale framework for low-light image enhancement using spatial–spectral information.” Computers & Electrical Engineering, 106, 108608, 2023. Publisher
  2. Yichi Zhang, Hengyu Liu, Dandan Ding, Zhan Ma. “Low Light RAW Image Enhancement Using Paired Fast Fourier Convolution and Transformer.” IEEE VCIP, 2022. IEEE

Research Experience

Hangzhou Normal University · Low-Light Image Enhancement

Undergraduate Research · Supervisor: Dandan Ding

2021.07 – 2022.06
  • Model experimentation: Contributed code changes and training experiments to integrate Fast Fourier Convolution (FFC) into RAW-to-RGB low-light enhancement models.
  • Experimental analysis: Collated experimental results and compared model variants using PSNR, SSIM and parameter counts; prepared architecture diagrams and group-meeting discussions.
  • Design and patent drafting: Proposed quality-based binary classification to select images for re-enhancement; drafted the patent application text, revised by patent counsel and published as CN114565817A in 2022 (second-listed inventor).

Relevant Industry Experience

Hangzhou Edge Exploration Technology Co., Ltd.

Algorithm Engineer · Sole algorithm engineer

2026.02 – 2026.07
  • Generative model development: Implemented symbolic-music generation models and PyTorch training pipelines; conducted model experiments and inference checks.
  • Training systems: Adapted model fine-tuning workflows for distributed training, checkpoint handling and GPU-memory diagnostics.

IM Motors Projects

Algorithm R&D · Vision Deployment & Autonomous-driving Evaluation

2024.04 – 2024.12
  • QNN deployment: Quantized and deployed a YOLOv5 vehicle-camera detector on Qualcomm SA8295 using ONNX and QNN; implemented C++ post-processing.
  • UniAD reproduction: Prepared a nuScenes subset and configured experiments on eight A100 GPUs; ran tracking, motion, occupancy and planning evaluations to establish a reproduction baseline.

ArcVideo Technology Co., Ltd.

Image Algorithm Engineer · Internship through Jun 2023

2023.02 – 2024.01
  • CUDA image processing: Implemented CUDA integral/squared-integral image kernels for regional mean/variance queries in atmospheric-light estimation, and guided filtering for transmission refinement. Integrated buffer reuse and stream-based execution into a video-processing pipeline.
  • Edge deployment: Adapted vision models to Jetson NX and RK3588 through TensorRT plugin integration, CUDA post-processing, RKNN conversion and image preprocessing.
  • Video frame interpolation: Developed and evaluated video frame interpolation using existing optical-flow components; integrated the model into a product beta.

Selected Open-source Work

  • Guided Filter using CUDA: GPU guided filtering for edge-preserving processing. Published development measurements: approximately 9.8 ms / 1080p image on RTX 4090, including allocation and memory copies, excluding the first warm-up run.
  • Integral Image using CUDA: Extended NVIDIA’s shuffle-scan sample for 16-bit input and squared-integral images, using warp-level prefix sums and wider accumulators.
  • MAA / MaaAI: Fine-tuned a game-domain Chinese PP-OCRv3 model (MAA #5895); refactored skill-readiness training, export and validation (MaaAI #8); contributed an updated MobileNetV4 classifier and expanded data (MAA #12490).

Published Patent Applications

  • Atmospheric Light Estimation for Dehazing Using Integral-Image Optimization. First-listed inventor · Chinese invention patent application CN117876256A · Published 2024-04-12.
  • Image Dehazing by Refining Atmospheric-Light Masks and Transmission Maps with Depth Information. First-listed inventor · Chinese invention patent application CN117893443A · Published 2024-04-16.

Technical Skills

  • Programming: Python · C++ · CUDA C/C++ · Linux · Git
  • ML / Deployment: PyTorch · DDP/FSDP2 · ONNX · TensorRT · QNN · RKNN