MS-COCO
Benchmarks
TaskDatasetSOTA resultTrendResults
MS COCO Karpathy (test)
149.1CIDEr
706
MS COCO (test-dev)
78.5mAP@.5
677
MS-COCO 2017 (val)
74mAP
264
MS-COCO 5K (test)
68.3R@1
244
MS COCO (val)
0.603mAP
242
MS-COCO
67R@1
235
MS-COCO 5K (test)
67.2R@1
217
MS-COCO (val)
1.53FID
215
MS-COCO (POPE Adversarial)
87.83Accuracy
205
MS-COCO
4.52FID
193
MS-COCO 5K (test)
84.8R@1
182
MS-COCO
100R@1
174
MS-COCO
80.5R@1
168
MS-COCO POPE (Popular)
90.76Accuracy
158
MS-COCO 2014 (val)
2.47FID
143
MS COCO novel classes
2,450nAP
132
MS-COCO 2017 (val)
19.259FID
131
MS-COCO 1K (test)
80.1R@1
128
MS COCO
47.9mAP Mask
123
MS COCO novel classes 2017 (val)
22.73AP
123
MS-COCO POPE Random
92.36Accuracy
121
MS-COCO 1K (test)
82R@1
121
MS COCO (test)
140.4CIDEr
120
MS-COCO (test)
84.08MAP
98
MS-COCO
83.8mAP
87