AI Paper Atlas

Research map / Vision-language and multimodal / Human and embodied modeling

RADIO1D: Elastic Representations for Condensed Vision Modeling

Greg Heinrich, Mike Ranzinger, Collin McCarthy, Natan Bagrov, Eugene Khvedchenya, Bryan Catanzaro, Jan Kautz, Andrew Tao, Pavlo Molchanov

ICML 2026 · Vision-language and multimodal › Human and embodied modeling · approach: Vision-language pretraining · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
fine-tune the vision encoder inside a VLMC-RADIOv4-H · MS-COCO CKACKAoff (spatial coherence): 0.281 to 0.035Hurt
use first-token versus last-token of the 1D sequence (ablation)RADIO1D · ADE20K linear probemIoU: 8.05 -> 40.23Improved
use first 16 tokens rather than last 16 (ablation)RADIO1D · ADE20K linear probemIoU: 20.84 -> 54.11Improved
add integrated learnable patch merging in the encoder (ablation)RADIO1D · VLM short run, DocVQA/InfoVQA/LongVideoBenchaccuracy: InfoVQA 69.39 -> 73.43; LongVideoBench 53.00 -> 56.54Improved
apply 2x2 pixel unshuffle on the 1D token sequence output (ablation)RADIO1D · RADIO1D trainingtraining convergenceHurt
shuffle 1D token order before 1D unshuffling (ablation)RADIO1D · RADIO1D trainingtraining lossNo clear effect
use triangular instead of uniform sampling of token length (ablation)RADIO1D (SO400M, short schedule) · ADE20K linear probemIoUImproved
place downscaling after block 24 (kappa=24) (ablation)RADIO1D (SO400M, short schedule) · ADE20K linear probemIoUImproved
use expansion factor 4 instead of 2 in downscaling (ablation)RADIO1D · parameter count and throughputparameters / throughputHurt
learn a 1D elastic sequence of 128 tokens instead of pruning C-RADIOv4 post-hoc with ToMeVLM with RADIO1D · 10 VLM benchmarks average, 128 tokensaverage accuracy: 70.31 -> 71.64Improved
apply CDPruner on top of the full 256-token RADIO1D instead of native elastic length (ablation)RADIO1D · 10 VLM benchmarks average, 128 tokensaverage accuracy: 71.64 -> 68.56Hurt
reduce visual tokens to 1 or fewer than 128 in RADIO1D (ablation)RADIO1D VLM · OCR benchmarks (DocVQA)accuracy: DocVQA 93.2 -> 51.1 (256 -> 1 token)Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy