Researchers from Nokia, Aalto University, and Tampere University have introduced CASR, a content-adaptive super-resolution post-filter for VVC that utilizes LoRA to optimize video quality. The method achieves significant BD-rate savings across luma and chroma components while maintaining compatibility with the VSEI NNPF standard.
This development addresses a critical weakness in neural super-resolution where standard AI models often improve luma at the expense of color fidelity. By integrating the upsampling operation directly into a VSEI-compliant model, Nokia enables a standardized path for decoder-side enhancement that does not require modifying the core VVC bitstream. This approach allows streaming providers to deliver high-resolution experiences from lower-resolution encodes, significantly reducing distribution costs without sacrificing visual quality. The industry should monitor the adoption of these NNPF-compliant filters in commercial decoders to see if LoRA becomes the preferred method for per-sequence optimization.
Nokia has moved beyond academic research into public software for neural network post-filtering. The company released open-source NNPF software that combines a VVC decoder path with support for NNPF SEI messages and real-time inference through OpenVINO, demonstrating how AI-enhanced video can be delivered in a standards-aligned, interoperable way rather than as a closed demo. The release builds on Nokia's earlier work within the VVC and VSEI standards ecosystems, showing how AI-based post-filtering can restore detail and sharpen textures without requiring increased bitrate.
Nokia's CASR research fits within a broader standardization push involving multiple industry players. Ericsson, Nokia, and Fraunhofer Heinrich Hertz Institute partnered in October 2025 to drive next-generation video coding standardization for immersive media and mobile video experiences, positioning VVC and its extensions as foundational for 6G-era delivery. This collaboration signals that major telecom equipment vendors view AI-enhanced video coding as a strategic priority for future network generations.
The CASR framework relies on the MPEG Neural Network Compression and Representation (NNR) standard to compress LoRA weight updates for transmission as side information. NNR combines sparsification, pruning, low-rank decomposition, quantization techniques, and DeepCABAC entropy coding to achieve up to 97% compression efficiency while maintaining inference accuracy, making per-sequence adaptation economically viable for streaming deployments. The VSEI standard (ISO/IEC 23002-7, ITU-T H.274) defines the SEI messages that allow neural network post-filters and associated parameter updates to be conveyed to decoders without modifying the core VVC decoding process, enabling backward-compatible deployment across VVC, HEVC, and AVC systems.
Nokia researchers have developed a content-adaptive super-resolution (CASR) framework for VVC video coding using Low-Rank Adaptation (LoRA). This method achieves significant BD-rate savings of up to 24.41% for chroma components. By utilizing the VSEI NNPF standard, this approach enables high-resolution video delivery without requiring modifications to the core VVC bitstream.
The CASR framework achieves significant BD-rate savings, specifically 10.93% for luma and 24.41% for chroma, while improving video quality through neural super-resolution without needing to modify the core VVC bitstream.
CASR utilizes Low-Rank Adaptation (LoRA) to fine-tune lightweight matrices at the encoder side while keeping base convolution kernels frozen, allowing for efficient per-sequence optimization.
Yes, the framework is fully compatible with the VSEI NNPF standard (ISO/IEC 23002-7, ITU-T H.274), allowing neural network post-filters to be conveyed to decoders without altering the core VVC decoding process.
The MPEG Neural Network Compression and Representation (NNR) standard is used to compress LoRA weight updates for transmission as side information, achieving up to 97% compression efficiency.
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source
