Object Detection2014en

Microsoft COCO: Common Objects in Context

330K images, 1.5 million object instances, 80 object categories. Standard benchmark for object detection and segmentation.

Samples:330,000
Metrics:mAP, AP50, AP75
Paper / WebsiteDownload
Current State of the Art

Co-DETR (Swin-L)

Research

66

mAP

mAP Progress Over Time

Showing 3 breakthroughs from Nov 2019 to Nov 2022

54.057.360.663.867.1Nov 2019May 2021Nov 2022mAPDate

Key Milestones

Nov 2019
EfficientDet-D7x

EfficientDet: Scalable and Efficient Object Detection.

55.1
Source
Mar 2022
DINO (Swin-L)

DINO: DETR with Improved DeNoising Anchor Boxes.

63.3
+14.9%
Source
Nov 2022
Co-DETR (Swin-L)Current SOTA

Co-DETR with Swin-L backbone. Current SOTA range for object detection.

66.0
+4.3%
Source
Total Improvement
19.8%
Time Span
3y 1m
Breakthroughs
3
Current SOTA
66.0

Top Models Performance Comparison

Top 5 models ranked by mAP

mAP1Co-DETR (Swin-L)66.0100.0%2InternImage-H65.499.1%3DINO (Swin-L)63.395.9%4YOLOv10-X57.487.0%5EfficientDet-D7x55.183.5%0%25%50%75%100%% of best
Best Score
66.0
Top Model
Co-DETR (Swin-L)
Models Compared
5
Score Range
10.9

mAPPrimary

#ModelScorePaper / CodeDate
1
Co-DETR (Swin-L)Open Source
Research		66
arxiv-paper
Dec 2025
2
InternImage-HOpen Source
Shanghai AI Lab		65.4
arxiv-paper
Dec 2025
3
DINO (Swin-L)Open Source
Research		63.3
arxiv-paper
Dec 2025
4
YOLOv10-XOpen Source
Tsinghua		57.4
github-readme
Dec 2025
5
EfficientDet-D7xOpen Source
Google		55.1
google-research
Dec 2025

Other Object Detection Datasets

Pascal VOC 2012
COCO Benchmark - Object Detection | CodeSOTA