AI Paper Atlas

Research map / Detection and segmentation / CNN features

Dynamic DETR: End-to-End Object Detection With Dynamic Attention

Xiyang Dai, Yinpeng Chen, Jianwei Yang, Pengchuan Zhang, Lu Yuan, Lei Zhang

ICCV 2021 · Detection and segmentation › CNN features · approach: Region-based and query-based detectors · 472 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 dynamic encoder (ablation)DETR · COCO val2017 (1x ablation)AP: 15.5 -> 24.1Improved
replace Deformable DETR encoder with dynamic encoder (ablation)Deformable DETR · COCO val2017 (1x ablation)AP: 37.2 -> 34.3Hurt
add dynamic encoder to dynamic decoder (ablation)Dynamic DETR · COCO val2017 (1x ablation)AP: 40.2 -> 42.9Improved
use dynamic decoder instead of Deformable DETR decoder (ablation)Deformable DETR · COCO val2017 (1x ablation)AP: 37.2 -> 40.2Improved
stack 6 dynamic encoders instead of 2 (ablation)Dynamic DETR · COCO val2017 (1x ablation)AP: 41.2 -> 42.9Improved
stack 6 dynamic decoders instead of 2 (ablation)Dynamic DETR · COCO val2017 (1x ablation)AP: 29.2 -> 42.9Improved
use 4 or 8 decoder heads instead of 2 (ablation)Dynamic DETR · COCO val2017 (1x ablation)AP: 42.5 -> 42.9No clear effect
increase decoder dimension from 512 to 2048 (ablation)Dynamic DETR · COCO val2017 (1x ablation)AP: 38.1 -> 42.9Improved
use 300 queries instead of 100 (ablation)Dynamic DETR · COCO val2017 (1x ablation)AP: 41.1 -> 42.9Improved
Dynamic DETR vs Deformable DETR convergenceDynamic DETR · COCO val2017 (extended training)AP: 47.2 mAP vs 45.3 mAPImproved
Dynamic DETR vs best traditional detector under 1x scheduleDynamic DETR · COCO val2017 (1x)AP: 41.4 -> 42.9Improved

Later papers that compared with this method

2 later accepted papers report Dynamic-DETR as a baseline in their results tables; 2 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 7 of 7 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