analiza aiqbench
This paper introduces Omni-Encoder, a unified multimodal architecture that jointly optimizes image, video, audio, and audio-visual data within a single shared SigLIP-2 Vision Transformer. By replacing post-hoc binding with end-to-end joint training and alternating-modality gradient accumulation, the approach demonstrates that fully shared parameters can outperform modality-specific encoders across 16 benchmarks. The work advances the state of contrastive pretraining by proving the efficacy of a unified spatiotemporal embedding space for diverse signal types.
When Does Sharing Work? Scaling a Shared SigLIP2 Across Image, Video, Audio, and Audio-Visual Representations
arXiv:2611.05406v1 Announce Type: cross Abstract: Contrastive pretraining has advanced image-text understanding, yet image, video, and audio encoders are still trained independently, with cross-modal alignment realized mainly through post-hoc binding to a frozen anchor space. We present Omni-Encoder, a general multimodal encoder that processes image, video, audio, and audio-visual inputs within a single SigLIP-2 Vision Transformer with fully shared parameters. Unlike anchor-centric binding or per-modality adapters, all modalities are optimized jointly end-to-end, and the framework is organized from input to training. A modality-native input stage first maps each signal into the shared embedding space, including a 1D-convolutional audio tokenizer that converts mel-spectrograms into tokens aligned with visual patches. A multimodal rotary position embedding parameterized by absolute timestamps then places visual frames and audio segments in a unified spatiotemporal coordinate system. Finally, alternating-modality training interleaves image, video, audio, and temporally aligned audio-visual batches and accumulates their gradients into one update, which we find central to stable joint optimization. Omni-Encoder achieves competitive results on 16 benchmarks across all four modalities with one shared parameter set, and joint training improves each modality over its separately trained counterpart. Code and models will be publicly released.
Title:When Does Sharing Work? Scaling a Shared SigLIP2 Across Image, Video, Audio, and Audio-Visual Representations
Abstract:Contrastive pretraining has advanced image-text understanding, yet image, video, and audio encoders are still trained independently, with cross-modal alignment realized mainly through post-hoc binding to a frozen anchor space. We present Omni-Encoder, a general multimodal encoder that processes image, video, audio, and audio-visual inputs within a single SigLIP-2 Vision Transformer with fully shared parameters. Unlike anchor-centric binding or per-modality adapters, all modalities are optimized jointly end-to-end, and the framework is organized from input to training. A modality-native input stage first maps each signal into the shared embedding space, including a 1D-convolutional audio tokenizer that converts mel-spectrograms into tokens aligned with visual patches. A multimodal rotary position embedding parameterized by absolute timestamps then places visual frames and audio segments in a unified spatiotemporal coordinate system. Finally, alternating-modality training interleaves image, video, audio, and temporally aligned audio-visual batches and accumulates their gradients into one update, which we find central to stable joint optimization. Omni-Encoder achieves competitive results on 16 benchmarks across all four modalities with one shared parameter set, and joint training improves each modality over its separately trained counterpart. Code and models will be publicly released.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
arXiv:2611.05406 [cs.CV]
(or arXiv:2611.05406v1 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2611.05406 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 298 cuvinte
Subiecte conexe
Sequential Contextual Fit Predicts Human Behavioural and Neural Dynamics Across Domains
arXiv · 18 sept. 2026
Constrained Hebbian Learning Supports Efficient Representational Allocation under Structural Constraints
arXiv · 20 iul. 2026
Why Is Video Still So Expensive? A Survey of Inference-Efficiency Mechanisms in Video and Audiovisual LLMs
arXiv · 10 sept. 2026
Syn-Omni: Structured Specialization and Progressive Collaboration for Omnimodal Embeddings
arXiv · 10 oct. 2026