Awesome
Deformable ConvNets is initially described in an arxiv tech report.
R-FCN is initially described in a NIPS 2016 paper.
Soft-NMS is initially described in an arxiv tech report.
Our goal was to test Soft-NMS with a state-of-the-art detector, so Deformable-R-FCN was trained on 800x1200 size images with 15 anchors. Multi-Scale testing was also added with 6 scales. Union of all boxes at each scale was computed before performing NMS. Please note that the repository does not include the scripts for multi-scale testing as I just cache the boxes for each different scale and do NMS separately. The scales used in multi-scale testing were as follows, [(480, 800), (576,900), (688, 1100), (800,1200), (1200, 1600), (1400, 2000)].
The trained model can be downloaded from here.
| <sub>training data</sub> | <sub>testing data</sub> | <sub>mAP</sub> | <sub>mAP@0.5</sub> | <sub>mAP@0.75</sub> | <sub>mAP@S</sub> | <sub>mAP@M</sub> | <sub>mAP@L</sub> | <sub>Recall</sub> | |
|---|---|---|---|---|---|---|---|---|---|
| <sub>Baseline D-R-FCN</sub> | <sub>coco trainval</sub> | <sub>coco test-dev</sub> | 35.7 | 56.8 | 38.3 | 15.2 | 38.8 | 51.5 | |
| <sub>D-R-FCN, ResNet-v1-101, NMS</sub> | <sub>coco trainval</sub> | <sub>coco test-dev</sub> | 37.4 | 59.6 | 40.2 | 17.8 | 40.6 | 51.4 | 48.3 |
| <sub>D-R-FCN, ResNet-v1-101, SNMS</sub> | <sub>coco trainval</sub> | <sub>coco test-dev</sub> | 38.4 | 60.1 | 41.6 | 18.5 | 41.6 | 52.5 | 53.8 |
| <sub>D-R-FCN, ResNet-v1-101, MST, NMS</sub> | <sub>coco trainval</sub> | <sub>coco test-dev</sub> | 39.8 | 62.4 | 43.3 | 22.6 | 42.3 | 52.2 | 52.9 |
| <sub>D-R-FCN, ResNet-v1-101, MST, SNMS</sub> | <sub>coco trainval</sub> | <sub>coco test-dev</sub> | 40.9 | 62.8 | 45.0 | 23.3 | 43.6 | 53.3 | 60.4 |