MS-COCO
Benchmarks
TaskDatasetSOTA resultTrendResults
MS COCO
35.9Recall
2
MS-COCO Corrupt
30Recall@1
2
MS-COCO Corrupt
37.2Recall@1
2
MS COCO 2017 (test)
43.4Image ARI
2
MS-COCO Karpathy Fine-tuned (test)
41.6BLEU-4
2
MS COCO subset of 20 categories 2017 (val)
95.24Explanation Accuracy
2
MS COCO subset of 24,237 images 2017 (train)
98.77Explanation Accuracy
2
MS-COCO (val)
16.05FID (Moongate)
2
MS-COCO 2017 (test)
25.5FID
2
MS-COCO 2017 (test)
43.4FID
2
MS COCO 2018 (test-dev)
35.9AP
2
MS-COCO novel categories (test)
16.8AP50
2
MS-COCO categories (train)
38.4mAP50
2
MS-COCO novel categories (test)
18.5mAP50
2
MS-COCO categories (train)
41.3mAP@0.50
2
MS COCO (val)
49.2mAP
2
MS-COCO
0.317BLEU-4
1
MS-COCO in-domain
32.36PSNR (dB)
1
MS-COCO filtered 256x256 (val)
12.89FID
1
MS-COCO 2017 (test)
13.6FID
1
MS-COCO randomly drawn boxes
50mAP50
1
MS-COCO randomly drawn boxes
1.2mAP50
1
MS-COCO 80 classes total (20 novel) (val)
—AP
0
MS-COCO gFSOD (test)
—AP (Overall)
0
MS-COCO 2014 (train)
—Primary metric
0