COCO
Benchmarks
TaskDatasetSOTA resultTrendResults
COCO-A (test)
82.9mIoU
6
COCO-LT v1.0 (test)
24.4AP
6
Real-world COCO-based evaluation settings Stage 1 Discovery Precision@10 > 0.6
69.8Image Overall Score
6
COCO Human Prompts (val)
21.36HPS
6
COCO Average across splits 2017 (Avg)
59.6bAP50
6
COCO 2017 (Split-4)
60.7bAP50
6
COCO 2017 (Split-3)
61.3Base AP50
6
COCO 2017 (Split-2)
57bAP50
6
COCO 2017 (Split-1)
59.4Base AP50
6
COCO 2014
11.31FID
6
COCO 2014
18.48FID
6
COCO-UniFS (val)
22.1AP
6
COCO 80-70 setting 2017 (test)
45AP (Recall-based)
6
COCO 80-40 setting 2017 (test)
44.8APr
6
COCO 80-5 setting 2017 (test)
45APr
6
COCO 80-1 setting 2017 (test)
44.5APr
6
COCO-Paragraph
75.1cIoU
6
COCO-cap
69.2mIoU
6
COCO 5,000 samples 2017 (val)
21.02FID
6
COCO Spatial 27K Vertical
74.1Accuracy (All)
6
COCO
19.5IoU
6
COCO-MIG
67.7Instance Success Rate (L2)
6
COCO 2017
13.82FID
6
COCO
51AP
6
COCO-80 (val)
51.8AP (Overall)
6