RADIO1D: Elastic Representations for Condensed Vision Modeling
For researchers and practitioners in vision-language modeling, RADIO1D offers a more efficient alternative to fixed patch-based 2D vision features, enabling flexible tradeoffs between accuracy and computational cost.
This paper introduces RADIO1D, a method that compresses images into a compact, variable-length 1D token sequence, enabling flexible accuracy-efficiency tradeoffs in vision-language models. It achieves competitive performance on multimodal benchmarks with lower computational overhead and better accuracy.
This paper challenges the assumption that vision-language models (VLMs) require fixed patch-based 2D vision features. Analyzing fine-tuned vision encoders, we find that representations become increasingly abstract and less spatially coherent during VLM training. Notably, models trained with image-text alignment (such as SigLIP2) develop a small number of specialized tokens that effectively summarize global image content. Building on this, we introduce RADIO1D, which compresses images into a compact, variable-length 1D token sequence using multi-teacher knowledge distillation and an autoencoder design. The resulting representations exhibit strong hierarchical summarization, enabling accurate scene understanding - even with a single token - and support improved composition-aware image retrieval. In VLMs, RADIO1D provides flexible accuracy-efficiency tradeoffs through adjustable token counts, delivering competitive performance on diverse multimodal benchmarks with lower computational overhead and better accuracy.