DMVC: Multi-Camera Video Compression Network aimed at Improving Deep Learning Accuracy

We introduce a cutting-edge video compression framework tailored for the age of ubiquitous video data, uniquely designed to serve machine learning applications. Unlike traditional compression methods that prioritize human visual perception, our innovative approach focuses on preserving semantic information critical for deep learning accuracy, while efficiently reducing data size. The framework operates on a batch basis, capable of handling multiple video streams simultaneously, thereby enhancing scalability and processing efficiency. It features a dual reconstruction mode: lightweight for real-time applications requiring swift responses, and high-precision for scenarios where accuracy is crucial. Based on a designed deep learning algorithms, it adeptly segregates essential information from redundancy, ensuring machine learning tasks are fed with data of the highest relevance. Our experimental results, derived from diverse datasets including urban surveillance and autonomous vehicle navigation, showcase DMVC's superiority in maintaining or improving machine learning task accuracy, while achieving significant data compression. This breakthrough paves the way for smarter, scalable video analysis systems, promising immense potential across various applications from smart city infrastructure to autonomous systems, establishing a new benchmark for integrating video compression with machine learning.

Predicting total time to compress a video corpus using online inference systems

Predicting the computational cost of compressing/transcoding clips in a video corpus is important for resource management of cloud services and VOD (Video On Demand) providers. Currently, customers of cloud video services are unaware of the cost of transcoding their files until the task is completed. Previous work concentrated on predicting perclip compression time, and thus estimating the cost of video compression. In this work, we propose new Machine Learning (ML) systems which predict cost for the entire corpus instead. This is a more appropriate goal since users are not interested in per-clip cost but instead the cost for the whole corpus. In this work, we evaluate our systems with respect to two video codecs (x264, x265) and a novel high-quality video corpus. We find that the accuracy of aggregate time prediction for a video corpus more than two times better than using per-clip predictions. Furthermore, we present an online inference framework in which we update the ML models as files are processed. A consideration of video compute overhead and appropriate choice of ML predictor for each fraction of corpus completed yields a prediction error of less than 5%. This is approximately two times better than previous work which proposed generalised predictors.

LongVU: Spatiotemporal Adaptive Compression for Long Video-Language Understanding

Multimodal Large Language Models (MLLMs) have shown promising progress in understanding and analyzing video content. However, processing long videos remains a significant challenge constrained by LLM's context size. To address this limitation, we propose LongVU, a spatiotemporal adaptive compression mechanism thats reduces the number of video tokens while preserving visual details of long videos. Our idea is based on leveraging cross-modal query and inter-frame dependencies to adaptively reduce temporal and spatial redundancy in videos. Specifically, we leverage DINOv2 features to remove redundant frames that exhibit high similarity. Then we utilize text-guided cross-modal query for selective frame feature reduction. Further, we perform spatial token reduction across frames based on their temporal dependencies. Our adaptive compression strategy effectively processes a large number of frames with little visual information loss within given context length. Our LongVU consistently surpass existing methods across a variety of video understanding benchmarks, especially on hour-long video understanding tasks such as VideoMME and MLVU. Given a light-weight LLM, our LongVU also scales effectively into a smaller size with state-of-the-art video understanding performance.

Standardizing Generative Face Video Compression using Supplemental Enhancement Information

This paper proposes a Generative Face Video Compression (GFVC) approach using Supplemental Enhancement Information (SEI), where a series of compact spatial and temporal representations of a face video signal (i.e., 2D/3D key-points, facial semantics and compact features) can be coded using SEI message and inserted into the coded video bitstream. At the time of writing, the proposed GFVC approach is an official "technology under consideration" (TuC) for standardization by the Joint Video Experts Team (JVET) of ISO/IEC JVT 1/SC 29 and ITU-T SG16. To the best of the authors' knowledge, the JVET work on the proposed SEI-based GFVC approach is the first standardization activity for generative video compression. The proposed SEI approach has not only advanced the reconstruction quality of early-day Model-Based Coding (MBC) via the state-of-the-art generative technique, but also established a new SEI definition for future GFVC applications and deployment. Experimental results illustrate that the proposed SEI-based GFVC approach can achieve remarkable rate-distortion performance compared with the latest Versatile Video Coding (VVC) standard, whilst also potentially enabling a wide variety of functionalities including user-specified animation/filtering and metaverse-related applications.

MambaSCI: Efficient Mamba-UNet for Quad-Bayer Patterned Video Snapshot Compressive Imaging

Color video snapshot compressive imaging (SCI) employs computational imaging techniques to capture multiple sequential video frames in a single Bayer-patterned measurement. With the increasing popularity of quad-Bayer pattern in mainstream smartphone cameras for capturing high-resolution videos, mobile photography has become more accessible to a wider audience. However, existing color video SCI reconstruction algorithms are designed based on the traditional Bayer pattern. When applied to videos captured by quad-Bayer cameras, these algorithms often result in color distortion and ineffective demosaicing, rendering them impractical for primary equipment. To address this challenge, we propose the MambaSCI method, which leverages the Mamba and UNet architectures for efficient reconstruction of quad-Bayer patterned color video SCI. To the best of our knowledge, our work presents the first algorithm for quad-Bayer patterned SCI reconstruction, and also the initial application of the Mamba model to this task. Specifically, we customize Residual-Mamba-Blocks, which residually connect the Spatial-Temporal Mamba (STMamba), Edge-Detail-Reconstruction (EDR) module, and Channel Attention (CA) module. Respectively, STMamba is used to model long-range spatial-temporal dependencies with linear complexity, EDR is for better edge-detail reconstruction, and CA is used to compensate for the missing channel information interaction in Mamba model. Experiments demonstrate that MambaSCI surpasses state-of-the-art methods with lower computational and memory costs. PyTorch style pseudo-code for the core modules is provided in the supplementary materials.

VidCompress: Memory-Enhanced Temporal Compression for Video Understanding in Large Language Models

Video-based multimodal large language models (Video-LLMs) possess significant potential for video understanding tasks. However, most Video-LLMs treat videos as a sequential set of individual frames, which results in insufficient temporal-spatial interaction that hinders fine-grained comprehension and difficulty in processing longer videos due to limited visual token capacity. To address these challenges, we propose VidCompress, a novel Video-LLM featuring memory-enhanced temporal compression. VidCompress employs a dual-compressor approach: a memory-enhanced compressor captures both short-term and long-term temporal relationships in videos and compresses the visual tokens using a multiscale transformer with a memory-cache mechanism, while a text-perceived compressor generates condensed visual tokens by utilizing Q-Former and integrating temporal contexts into query embeddings with cross attention. Experiments on several VideoQA datasets and comprehensive benchmarks demonstrate that VidCompress efficiently models complex temporal-spatial relations and significantly outperforms existing Video-LLMs.

Generative Human Video Compression with Multi-granularity Temporal Trajectory Factorization

In this paper, we propose a novel Multi-granularity Temporal Trajectory Factorization framework for generative human video compression, which holds great potential for bandwidth-constrained human-centric video communication. In particular, the proposed motion factorization strategy can facilitate to implicitly characterize the high-dimensional visual signal into compact motion vectors for representation compactness and further transform these vectors into a fine-grained field for motion expressibility. As such, the coded bit-stream can be entailed with enough visual motion information at the lowest representation cost. Meanwhile, a resolution-expandable generative module is developed with enhanced background stability, such that the proposed framework can be optimized towards higher reconstruction robustness and more flexible resolution adaptation. Experimental results show that proposed method outperforms latest generative models and the state-of-the-art video coding standard Versatile Video Coding (VVC) on both talking-face videos and moving-body videos in terms of both objective and subjective quality. The project page can be found at https://github.com/xyzysz/Extreme-Human-Video-Compression-with-MTTF.

ECVC: Exploiting Non-Local Correlations in Multiple Frames for Contextual Video Compression

In Learned Video Compression (LVC), improving inter prediction, such as enhancing temporal context mining and mitigating accumulated errors, is crucial for boosting rate-distortion performance. Existing LVCs mainly focus on mining the temporal movements within adjacent frames, neglecting non-local correlations among frames. Additionally, current contextual video compression models use a single reference frame, which is insufficient for handling complex movements. To address these issues, we propose leveraging non-local correlations across multiple frames to enhance temporal priors, significantly boosting rate-distortion performance. To mitigate error accumulation, we introduce a partial cascaded fine-tuning strategy that supports fine-tuning on full-length sequences with constrained computational resources. This method reduces the train-test mismatch in sequence lengths and significantly decreases accumulated errors. Based on the proposed techniques, we present a video compression scheme ECVC. Experiments demonstrate that our ECVC achieves state-of-the-art performance, reducing 7.3% and 10.5% more bit-rates than DCVC-DC and DCVC-FM over VTM-13.2 low delay B (LDB), respectively, when the intra period (IP) is 32. Additionally, ECVC reduces 11.1% more bit-rate than DCVC-FM over VTM-13.2 LDB when the IP is -1. Our Code will be available at https://github.com/JiangWeibeta/ECVC.

MEGA: Memory-Efficient 4D Gaussian Splatting for Dynamic Scenes

4D Gaussian Splatting (4DGS) has recently emerged as a promising technique for capturing complex dynamic 3D scenes with high fidelity. It utilizes a 4D Gaussian representation and a GPU-friendly rasterizer, enabling rapid rendering speeds. Despite its advantages, 4DGS faces significant challenges, notably the requirement of millions of 4D Gaussians, each with extensive associated attributes, leading to substantial memory and storage cost. This paper introduces a memory-efficient framework for 4DGS. We streamline the color attribute by decomposing it into a per-Gaussian direct color component with only 3 parameters and a shared lightweight alternating current color predictor. This approach eliminates the need for spherical harmonics coefficients, which typically involve up to 144 parameters in classic 4DGS, thereby creating a memory-efficient 4D Gaussian representation. Furthermore, we introduce an entropy-constrained Gaussian deformation technique that uses a deformation field to expand the action range of each Gaussian and integrates an opacity-based entropy loss to limit the number of Gaussians, thus forcing our model to use as few Gaussians as possible to fit a dynamic scene well. With simple half-precision storage and zip compression, our framework achieves a storage reduction by approximately 190$\times$ and 125$\times$ on the Technicolor and Neural 3D Video datasets, respectively, compared to the original 4DGS. Meanwhile, it maintains comparable rendering speeds and scene representation quality, setting a new standard in the field.

AsymKV: Enabling 1-Bit Quantization of KV Cache with Layer-Wise Asymmetric Quantization Configurations

Large language models have shown exceptional capabilities in a wide range of tasks, such as text generation and video generation, among others. However, due to their massive parameter count, these models often require substantial storage space, imposing significant constraints on the machines deploying LLMs. To overcome this limitation, one research direction proposes to compress the models using integer replacements for floating-point numbers, in a process known as Quantization. Some recent studies suggest quantizing the key and value cache (KV Cache) of LLMs, and designing quantization techniques that treat the key and value matrices equivalently. This work delves deeper into the asymmetric structural roles of KV Cache, a phenomenon where the transformer's output loss is more sensitive to the quantization of key matrices. We conduct a systematic examination of the attention output error resulting from key and value quantization. The phenomenon inspires us to propose an asymmetric quantization strategy. Our approach allows for 1-bit quantization of the KV cache by implementing distinct configurations for key and value matrices. We carry out experiments across a variety of datasets, demonstrating that our proposed model allows for the quantization of up to 75% decoder layers with 1 bit, while simultaneously maintaining performance levels comparable to those of the models with floating parameters.