Weakly-supervised instance segmentation 벤치마크
Weakly-supervised instance segmentation on PASCAL VOC 2012 val
Average Best Overlap
- 2020-01-30 — WSIS_CL: Average Best Overlap 48.2
- 2020-09-10 — LIID: Average Best Overlap 50.8
- 2021-03-16 — BBAM: Average Best Overlap 63.0
| Rank | Model | Average Best Overlap | mAP@0.25 | mAP@0.5 | mAP@0.75 | Paper | Code | Year |
|---|---|---|---|---|---|---|---|---|
| 1 | BBAM | 63.0 | 76.8 | 63.7 | 31.8 | BBAM: Bounding Box Attribution Map for Weakly Supervised Semantic and Instance Segmentation | jbeomlee93/BBAM | 2021 |
| 2 | LIID | 50.8 | - | 48.4 | 24.9 | Leveraging Instance-, Image- and Dataset-Level Information for Weakly Supervised Instance Segmentation | yun-liu/LIID | 2020 |
| 3 | WSIS_CL | 48.2 | 56.6 | 38.1 | 12.3 | Weakly Supervised Instance Segmentation by Deep Community Learning | 2020 | |
| 4 | CIM + Mask R-CNN | – | 68.7 | 55.9 | 30.9 | Complete Instances Mining for Weakly Supervised Instance Segmentation | ZechengLi19/CIM | 2024 |
| 5 | BESTIE (point label, proposal-free) | - | 66.4 | 56.1 | 30.2 | Beyond Semantic to Instance Segmentation: Weakly-Supervised Instance Segmentation via Semantic Knowledge Transfer and Self-Refinement | clovaai/BESTIE | 2021 |
| 6 | BESTIE (image-level label, proposal-free) | - | 61.2 | 51.0 | 26.6 | Beyond Semantic to Instance Segmentation: Weakly-Supervised Instance Segmentation via Semantic Knowledge Transfer and Self-Refinement | clovaai/BESTIE | 2021 |