Method Details
Details for method 'MRCNN_VSCMLab_ROB'
Method overview
| name | MRCNN_VSCMLab_ROB |
| challenge | instance-level semantic labeling |
| details | MaskRCNN+FPN with pre-trained COCO model. ms-training with short edge [800, 1024] inference with shore edge size 800 Randomly subsample ScanNet to the size close to CityScape optimizer: Adam learning rate: start from 1e-4 to 1e-3 with linear warm up schedule. decrease by factor of 0.1 at 200, 300 epoch. epoch: 400 step per epoch: 500 roi_per_im: 512 |
| publication | Anonymous |
| project page / code | |
| used Cityscapes data | fine annotations |
| used external data | ImageNet COCO KITTI ScanNet WildDash |
| runtime | 1 s P100 |
| subsampling | no |
| submission date | May, 2018 |
| previous submissions |
Average results
| Metric | Value |
|---|---|
| AP | 14.8465 |
| AP50% | 29.4643 |
| AP100m | 24.8178 |
| AP50m | 29.3188 |
Class results
| Class | AP | AP50% | AP100m | AP50m |
|---|---|---|---|---|
| person | 15.6718 | 37.7253 | 30.2901 | 30.9321 |
| rider | 11.5244 | 38.0798 | 19.6352 | 20.3297 |
| car | 36.721 | 57.666 | 57.6519 | 62.6666 |
| truck | 13.569 | 21.1247 | 22.4198 | 28.1372 |
| bus | 18.6891 | 29.3996 | 32.4247 | 40.9489 |
| train | 14.2741 | 26.6113 | 24.07 | 38.4145 |
| motorcycle | 8.32217 | 25.1081 | 12.0506 | 13.1213 |
| bicycle | 0 | 0 | 0 | 0 |