580 California St., Suite 400
San Francisco, CA, 94104
Key research themes
1. How can optimized quantization and entropy encoding improve compression efficiency without compromising video quality in modern video codecs?
This research theme focuses on enhancing video compression by designing and applying optimized quantization matrices combined with advanced entropy encoding techniques. Such improvements aim to achieve higher compression rates while preserving or even improving the decoded video quality, particularly relevant for standards like HEVC that incorporate quantization matrices as coding tools.
Key finding: The authors propose an entropy encoding method integrating an optimized quantization matrix (WE-OQM) within HEVC, achieving up to 35.29% better performance than standard entropy encoding and 62.5% better than weighted entropy... Read more
Key finding: This paper introduces an advanced context modeling scheme for coding quantized transform coefficients tailored for large transform blocks (up to 128x128), extending beyond the CABAC scheme in H.264/AVC. By adaptively... Read more
Key finding: The researchers propose DST-7 and DCT-8 transform approximations and fast computation algorithms leveraging the relationship with DCT-2, suitable for Versatile Video Coding (VVC). These transforms, combined with Multiple... Read more
2. What roles do advanced coding frameworks and adaptive partitioning techniques play in balancing compression efficiency, computational complexity, and real-time processing in next-generation video codecs?
This theme investigates novel codec architectures and control mechanisms, including distributed video coding, low complexity enhancement layers, and dynamic frame partitioning, to optimize the trade-offs between compression performance, computational cost, and latency for real-time or bandwidth-limited applications.
Key finding: This study proposes the QUAM scheme that selectively drops zero quantized transform blocks in distributed residual video coding (DRVC), reducing bit planes that require channel coding. This reduces encoder and decoder... Read more
Key finding: LCEVC, designed as a low-complexity enhancement layer over existing codecs (AVC, HEVC, EVC, VVC), shows bitrate savings up to 40% with adaptive streaming at full resolution. Verification tests employing objective and... Read more
Key finding: The authors introduce a dynamic Tile and Rectangular Slice (TRS) frame partitioning approach for VVC encoding that jointly optimizes multi-thread encoding time and encoding quality loss. Using spatial content and prior frame... Read more
3. How do emerging AI-driven and diffusion-based methods enhance perceptual quality and compression efficiency beyond classical video coding standards?
This theme explores the integration of deep learning, neural network architectures, and diffusion models in video compression, aiming to surpass traditional codec limitations by leveraging learned priors and advanced perceptual models to achieve improved compression ratios and visual perceptual quality.
Key finding: This paper presents a novel diffusion-based video compression approach that utilizes denoising diffusion generative models fine-tuned via low-rank adaptations (LoRA) to compress video frames with very low-quality guidance... Read more
Key finding: The authors design an end-to-end trainable deep learning compression network integrating optical flow estimation for motion compensation with convolutional and recurrent neural networks. Incorporating optical flow improves... Read more