AI Paper Atlas

Research map / Detection and segmentation / Open-set and few-shot detection

UniverSeg: Universal Medical Image Segmentation

Victor Ion Butoi, Jose Javier Gonzalez Ortiz, Tianyu Ma, Mert R. Sabuncu, John Guttag, Adrian V. Dalca

ICCV 2023 · Detection and segmentation › Open-set and few-shot detection · approach: Interactive and referring segmentation · 245 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
replace few-shot baselines (SENet, ALPNet, PANet) with UniverSeg CrossBlock networkUniverSeg · Held-out medical datasets (avg of 6)Dice: 50.1 -> 71.8 (vs best baseline SENet)Improved
add task augmentation (alone, on medical data) (ablation)UniverSeg · Held-out medical datasets (avg of 6)Dice: 62.7 -> 70.4Improved
add in-task augmentation (alone, on medical data) (ablation)UniverSeg · Held-out medical datasets (avg of 6)Dice: 62.7 -> 67.0Improved
add synthetic tasks to medical data with both augmentations (ablation)UniverSeg · Held-out medical datasets (avg of 6)Dice: 70.0 -> 71.8Improved
add synthetic tasks to medical tasks without augmentation (ablation)UniverSeg · Held-out medical datasets (avg of 6)Dice: 62.7 -> 64.5Improved
train only on synthetic tasks, no medical data (ablation)UniverSeg · Held-out medical datasets (avg of 6)Dice: 61.7 (comparable to 62.7 with medical only)Improved
add task diversity (augmentations) to training on STARE (ablation)UniverSeg · STAREDice: 63.8 -> 42.2Hurt
increase training support size from 1 to 16 (ablation)UniverSeg · Held-out medical datasets (avg of 6)Dice: 53.7 -> 69.9Improved
increase support size beyond 16 to 64 (ablation)UniverSeg · Held-out medical datasets (avg of 6)Dice: 69.9 -> 71Improved
ensemble predictions over multiple support sets (ablation)UniverSeg · Held-out medical datasets (avg of 6)Dice: +2.4-3.1 Dice for N < 16Improved
increase number of training datasets (choice of subset matters) (ablation)UniverSeg · Held-out medical datasets (avg of 6)Dice: best vs worst 10% subset gap: 17.3 DiceImproved
increase number of labeled support examples at inference (ablation)UniverSeg · WBC and PanDentalDiceImproved

Later papers that compared with this method

6 later accepted papers report UniverSeg as a baseline in their results tables; 6 of them report a better number on most of the results they share with it.

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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