AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Enhancing Vision Transformers for Object Detection via Context-Aware Token Selection and Packing

Tianyi Zhang, Baoxin Li, Jae-sun Seo, Yu Cao

ICLR 2026 · Architectures and efficiency › Attention and transformers · approach: Efficient and robust vision transformers

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
add SPA dynamic token selection instead of uniform top-50 (ablation)SPT-T · PASCAL VOC 2012mAP: 44.42 -> 44.49Improved
add select loss Lselect (ablation)SPT-T · PASCAL VOC 2012mAP: 44.49 -> 44.60Improved
add select loss Lselect (lowers mean select ratio) (ablation)SPT-T · PASCAL VOC 2012mean select ratio: 59.77 -> 29.60Improved
start SPA at the third stage rather than only the 4th stage (ablation)SPT-T · BDD100KAP: 21.9 -> 22.6Improved
start SPA at the third stage rather than the 2nd stage (ablation)SPT-T · BDD100KAP: 20.5 -> 22.6Improved
start SPA at the third stage rather than in all four stages (ablation)SPT-T · BDD100KAP: 18.3 -> 22.6Improved
weight positive tokens 5x in select loss (ablation)SPT · BDD-SmAP: 33% -> 71% select ratio, mAP dropped to 5.3Hurt
use SPT-T/16 (more blocks) rather than SPT-T/12 (ablation)SPT-T · BDD100KAP: 22.3 -> 22.6Improved
replace DynamicSwin-T backbone with SPT-TCascade Mask RCNN · COCO2017AP: 44.3 -> 47.1Improved
replace DynamicSwin-S backbone with SPT-SCascade Mask RCNN · COCO2017AP: 47.2 -> 49.3Improved
replace DynamicSwin-B backbone with SPT-BCascade Mask RCNN · COCO2017AP: 50.5 -> 53.2Improved
replace DynamicSwin-S backbone with SPT-S for instance segmentationMask RCNN · COCO2017 instance segmentationAP: 39.6 -> 40.9Improved

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