AI Paper Atlas

Research map / Detection and segmentation / CNN features

StageInteractor: Query-based Object Detector with Cross-stage Interaction

Yao Teng, Haisong Liu, Sheng Guo, Limin Wang

ICCV 2023 · Detection and segmentation › CNN features · approach: Efficient DETR query design · 14 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
add cross-stage label assignment (ablation)AdaMixer ResNet-50 100 queries · COCO minival 1xAP: 43.0 -> 44.8Improved
add cross-stage dynamic filter reuse (ablation)AdaMixer ResNet-50 100 queries · COCO minival 1xAP: 44.1 -> 44.8Improved
use cross-stage label assigner instead of hybrid layer matching (ablation)AdaMixer 100 queries · COCO minival 1xAP: 41.8 -> 44.8Improved
add IoU score to cross-stage label gathering (ablation)cross-stage label assigner · COCO minival 1xAP: 43.1 -> 44.8Improved
gather labels from all previous stages (alpha_i=1) instead of from i-1 (ablation)cross-stage label assigner · COCO minival 1xAP: 44.8 -> 43.3Hurt
reuse more spatial filters (3 instead of 1) (ablation)spatial dynamic mixing · COCO minival 1xAP: 44.8 -> 44.4Hurt
increase number of reused channel-mixing filters (ablation)channel mixing · COCO minival 1xAP: 44.1 -> 44.8Improved
fuse previous and current filters with adapters instead of using one directly (ablation)channel mixing · COCO minival 1xAP: 44.5 -> 44.8Improved
also modify localization supervision in the assigner (ablation)cross-stage label assigner · COCO minival 1xAP: 44.8 -> 44.8No clear effect
add cross-stage label assignment to DINODINO ResNet-50 12 epochs · COCO minivalAP: 49.2 -> 49.7Improved
add cross-stage label assignment to DINO with Swin-BDINO Swin-B · COCO minivalAP: 55.8 -> 56.2Improved
add cross-stage label assignment and filter reuseAdaMixer · COCO minival speedinference speed (img/s): 13.5 -> 11.6Hurt

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