notesum.ai
Published at November 26Self-supervised Video Instance Segmentation Can Boost Geographic Entity Alignment in Historical Maps
cs.CV
Released Date: November 26, 2024
Authors: Xue Xia1, Randall Balestriero2, Tao Zhang3, Lorenz Hurni1
Aff.: 1ETH Zurich; 2Brown University; 3Wuhan University

| Pretraining Configurations | AP | AP50 | AP75 | Precision | Recall | F1 |
| w/o Pretraining | 20.7 | 40.5 | 21.1 | 0.47 | 0.60 | 0.53 |
| COCO Images | 46.2 | 72.8 | 52.0 | 0.76 | 0.76 | 0.76 |
| YouTubeVIS-2019 | 44.8 | 69.8 | 49.8 | 0.74 | 0.76 | 0.75 |
| ImageNet Synthetic Videos | 32.0 | 56.5 | 33.5 | 0.64 | 0.68 | 0.66 |
| Map Synthetic Videos | 45.6 | 72.2 | 49.8 | 0.75 | 0.76 | 0.76 |