Search papers, labs, and topics across Lattice.
The paper introduces OneVision-Encoder (OV-Encoder), a novel vision architecture designed to align with the information-theoretic principles of video codecs by focusing computation on sparse, high-entropy regions. OV-Encoder uses Codec Patchification to process only 3.1%-25% of image regions and employs a shared 3D RoPE with a large-scale cluster discrimination objective to unify spatial and temporal reasoning. Experiments demonstrate that OV-Encoder outperforms strong vision backbones like Qwen3-ViT and SigLIP2 across various multimodal benchmarks, achieving a 4.1% improvement on video understanding while using fewer visual tokens and pretraining data.
Turns out, skipping the boring parts of a video (like static backgrounds) makes your vision AI both faster and smarter, beating state-of-the-art models with less data.
Hypothesis. Artificial general intelligence is, at its core, a compression problem. Effective compression demands resonance: deep learning scales best when its architecture aligns with the fundamental structure of the data. These are the fundamental principles. Yet, modern vision architectures have strayed from these truths: visual signals are highly redundant, while discriminative information, the surprise, is sparse. Current models process dense pixel grids uniformly, wasting vast compute on static background rather than focusing on the predictive residuals that define motion and meaning. We argue that to solve visual understanding, we must align our architectures with the information-theoretic principles of video, i.e., Codecs. Method. OneVision-Encoder encodes video by compressing predictive visual structure into semantic meaning. By adopting Codec Patchification, OV-Encoder abandons uniform computation to focus exclusively on the 3.1%-25% of regions rich in signal entropy. To unify spatial and temporal reasoning under irregular token layouts, OneVision-Encoder employs a shared 3D RoPE and is trained with a large-scale cluster discrimination objective over more than one million semantic concepts, jointly capturing object permanence and motion dynamics. Evidence. The results validate our core hypothesis: efficiency and accuracy are not a trade-off; they are positively correlated. When integrated into LLM, it consistently outperforms strong vision backbones such as Qwen3-ViT and SigLIP2 across 16 image, video, and document understanding benchmarks, despite using substantially fewer visual tokens and pretraining data. Notably, on video understanding tasks, OV-Encoder achieves an average improvement of 4.1% over Qwen3-ViT. Codec-aligned, patch-level sparsity is a foundational principle, enabling OV-Encoder as a scalable engine for next-generation visual generalists.
