Semantic Segmentation on COCO 2017 (test-dev)
38.5mAPHCL
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| HCLPre-train dataset=COCO, Baseline architecture=Dense-CL, Cropping strategy=selective search boxes2022.12 | 38.5 | 60.6 | 41.4 | |
| HCLPre-train dataset=OpenImages, Baseline architecture=Dense-CL, Cropping strategy=selective search boxes2022.12 | 38.3 | 59.4 | 40.6 | |
| Dense-CL (Baseline + bbox)Pre-train dataset=COCO, Baseline architecture=Dense-CL, Cropping strategy=selective search boxes2022.12 | 37.5 | 59.5 | 40.4 | |
| HCLPre-train dataset=COCO, Baseline architecture=ORL, Cropping strategy=selective search boxes2022.12 | 37.3 | 58.5 | 40 | |
| Dense-CL (Baseline + bbox)Pre-train dataset=OpenImages, Baseline architecture=Dense-CL, Cropping strategy=selective search boxes2022.12 | 37.2 | 58.3 | 39.7 | |
| HCLPre-train dataset=COCO, Baseline architecture=MoCo-v2, Cropping strategy=selective search boxes2022.12 | 37 | 58.3 | 39.7 | |
| ORLPre-train dataset=COCO, Baseline architecture=ORL, Cropping strategy=random crops2022.12 | 36.3 | 57.3 | 38.9 | |
| MoCo-v2 (Baseline + bbox)Pre-train dataset=COCO, Baseline architecture=MoCo-v2, Cropping strategy=selective search boxes2022.12 | 36 | 57.3 | 38.8 | |
| Dense-CLPre-train dataset=COCO, Baseline architecture=Dense-CL, Cropping strategy=random crops2022.12 | 35.7 | 56.5 | 38.4 | |
| MoCo-v2Pre-train dataset=COCO, Baseline architecture=MoCo-v2, Cropping strategy=random crops2022.12 | 34.8 | 55.3 | 37.3 | |
| Dense-CLPre-train dataset=OpenImages, Baseline architecture=Dense-CL, Cropping strategy=random crops2022.12 | 34.8 | 55.3 | 37.8 |