Chuqin Zhou（周楚骎）

Chuqin Zhou（周楚骎）

Master’s student at Shanghai Jiao Tong University

💼 Experience

ByteDance · Video Coding and Model Deployment Intern

Jun. 2026 – Aug. 2026

  • Developed a neural-network-based content-adaptive encoding (CAE) model to replace an XGBoost baseline, predicting VMAF from pre-encoding features to guide CRF selection. Built the feature extraction, data processing, training, and evaluation pipeline.

Ultra-Low-Bitrate HD Facial Video Coding · Huawei Collaboration

Sep. 2025 – Present

  • Contributed a reference-frame mechanism to a codebook-based face restoration model for a standards-compatible solution.
  • Led end-to-end work on loss-resilient neural reconstruction and video-tokenizer-based feature coding, targeting perceptual quality and temporal consistency at extremely low bitrates.
  • Outcomes include Every Packet Counts (ACM MM 2026, Oral, co-first author) and Efficient Generative Video Compression with Causal Latent Modeling (first author; submitted to AAAI 2027).

Generative Image Compression at Ultra-Low Bitrates

Sep. 2025 – Feb. 2026

  • Developed a dual-representation framework combining explicit semantics with implicit textures to address semantic drift and missing detail. Used latent features and text for semantic transmission, and reverse-channel coding of diffusion-trajectory noise for texture reconstruction.
  • Designed a plug-and-play distortion–perception control module. Achieved 29.9% bitrate savings at comparable quality in the reported evaluation, while preserving semantic consistency and realistic textures.
  • Outcome: Dual-Representation Image Compression at Ultra-Low Bitrates via Explicit Semantics and Implicit Textures (TCSVT 2026, first author).

Low-Bitrate Video Coding · Huawei Collaboration

Jul. 2024 – Jul. 2025

  • Contributed a differentiable VMAF implementation in PyTorch for neural video codec optimization, and led video-tokenizer feature coding for perceptual reconstruction.
  • Led the design and end-to-end training of lightweight pre- and post-processing networks for real-time, standards-compatible video enhancement on edge devices.
  • The proposed solution was ready for integration into Huawei MeeTime.
  • Outcomes include Differentiable VMAF (ISCAS 2025, second author) and Perceptual Neural Video Compression with Video Variational AutoEncoder at Low Bitrates.

Controllable Distortion–Perception Tradeoff for Neural Image Compression

Jul. 2024 – Dec. 2024

  • Developed a latent-diffusion-based feature transformation and adaptive fusion module at the decoder, enabling continuous distortion–perception control during inference without retraining the original codec.
  • Used an auxiliary encoder to provide perception-oriented training guidance, achieving a wide controllable reconstruction range while preserving the original codec’s compression capabilities.
  • Outcomes: Controllable Distortion-Perception Tradeoff Through Latent Diffusion for Neural Image Compression (AAAI 2025, first author) and the granted patent CN119211548B (first student inventor).

Huazhong University of Science and Technology · Research Intern

Jun. 2022 – Jun. 2023 · Wuhan, China