OmniZip: Learning a Unified and Lightweight Lossless Compressor for Multi-Modal Data
arXiv:2602.22286v1 Announce Type: new Abstract: Lossless compression is essential for efficient data storage and transmission. Although learning-based lossless compressors achieve strong results, most of them are designed for a single modality, leading to redundant compressor deployments in multi-modal settings. Designing a unified multi-modal compressor is critical yet challenging, as different data types vary largely in format, dimension, and statistics. Multi-modal large language models offer a promising resolution but remain too complex for practical use. Thus, we propose \textbf{OmniZip}, \textbf{a unified and lightweight lossless compressor for multi-modal data (like image, text, speech, tactile, database, and gene sequence)}. Built on a lightweight backbone, OmniZip incorporates three key components to enable efficient multi-modal lossless compression: a modality-unified tokenizer that reversibly transforms diverse data into tokens, a modality-routing context learning mechanism
arXiv:2602.22286v1 Announce Type: new Abstract: Lossless compression is essential for efficient data storage and transmission. Although learning-based lossless compressors achieve strong results, most of them are designed for a single modality, leading to redundant compressor deployments in multi-modal settings. Designing a unified multi-modal compressor is critical yet challenging, as different data types vary largely in format, dimension, and statistics. Multi-modal large language models offer a promising resolution but remain too complex for practical use. Thus, we propose \textbf{OmniZip}, \textbf{a unified and lightweight lossless compressor for multi-modal data (like image, text, speech, tactile, database, and gene sequence)}. Built on a lightweight backbone, OmniZip incorporates three key components to enable efficient multi-modal lossless compression: a modality-unified tokenizer that reversibly transforms diverse data into tokens, a modality-routing context learning mechanism that enables flexible multi-modal context modeling, and a modality-routing feedforward design that further enhances the model's nonlinear representation flexibility. A reparameterization training strategy is used to enhance model capacity. OmniZip outperforms or matches other state-of-the-art compressors on multiple modalities, achieving 42\%, 57\%, 62\% and 42\%, 53\% higher compression efficiency than gzip on CLIC-M, TouchandGo, enwik9, LibriSpeech, and WikiSQL datasets, respectively. It also supports near real-time inference on resource-constrained edge devices, reaching about 1MB/s on MacBook CPUs and iPhone NPUs. Our code is released at https://github.com/adminasmi/OmniZip-CVPR2026.
Executive Summary
This article proposes OmniZip, a unified and lightweight lossless compressor for multi-modal data that achieves state-of-the-art results in various modalities while supporting near real-time inference on resource-constrained edge devices. OmniZip incorporates three key components: a modality-unified tokenizer, a modality-routing context learning mechanism, and a modality-routing feedforward design. The model outperforms or matches other state-of-the-art compressors on multiple modalities, including image, text, speech, tactile, database, and gene sequence. The authors' use of a reparameterization training strategy enhances the model's capacity, enabling efficient compression and inference.
Key Points
- ▸ OmniZip is a unified and lightweight lossless compressor for multi-modal data.
- ▸ The model incorporates three key components: a modality-unified tokenizer, a modality-routing context learning mechanism, and a modality-routing feedforward design.
- ▸ OmniZip outperforms or matches other state-of-the-art compressors on multiple modalities.
Merits
Scalability
OmniZip's unified architecture enables efficient compression of diverse multi-modal data, making it a scalable solution for various applications.
Practicality
The model's ability to support near real-time inference on resource-constrained edge devices makes it a practical solution for real-world applications.
Flexibility
OmniZip's modality-routing design allows for flexible adaptation to different modalities and applications, increasing its flexibility and versatility.
Demerits
Complexity
The model's architecture may be complex and difficult to interpret, which could hinder its adoption and maintenance in certain applications.
Resource Requirements
While OmniZip supports inference on resource-constrained devices, its training requirements may still be computationally intensive and resource-hungry.
Expert Commentary
The proposal of OmniZip is a significant advancement in the field of data compression, as it provides a unified and lightweight solution for multi-modal data. While the model's architecture may be complex, its scalability and practicality make it an attractive solution for various applications. However, further research is needed to explore the model's limitations and potential biases. The implications of OmniZip are far-reaching, with potential applications in edge AI, IoT, and data management. As the field of deep learning for data compression continues to evolve, OmniZip is an important milestone that will likely shape the future of data storage and management.
Recommendations
- ✓ Further research is needed to explore the model's limitations and potential biases, as well as to develop more efficient training and inference strategies.
- ✓ The authors should provide more detailed analysis and evaluation of the model's performance on various datasets and applications, including a comparison with other state-of-the-art compressors.
