Details for method 'AdaptIS'
Method overview
| name |
AdaptIS |
| challenge |
instance-level semantic labeling |
| details |
Adaptive Instance Selection network architecture for class-agnostic instance segmentation. Given an input image and a point (x, y), it generates a mask for the object located at (x, y). The network adapts to the input point with a help of AdaIN layers, thus producing different masks for different objects on the same image. AdaptIS generates pixel-accurate object masks, therefore it accurately segments objects of complex shape or severely occluded ones. |
| publication |
Anonymous
|
| project page / code |
|
| used Cityscapes data |
fine annotations |
| used external data |
ImageNet |
| runtime |
n/a |
| subsampling |
no |
| submission date |
June, 2019 |
| previous submissions |
|
Average results
| Metric |
Value |
| AP | 32.4791 |
| AP50% | 52.5225 |
| AP100m | 48.2192 |
| AP50m | 52.103 |
Class results
| Class |
AP | AP50% | AP100m | AP50m |
| person | 31.3879 | 59.4919 | 49.7141 | 49.794 |
| rider | 29.0871 | 56.4277 | 45.907 | 46.7794 |
| car | 49.8044 | 75.1276 | 69.3574 | 71.3106 |
| truck | 31.6472 | 38.9581 | 45.6035 | 54.319 |
| bus | 41.6674 | 52.7572 | 64.9896 | 77.2432 |
| train | 39.4022 | 56.6251 | 58.0231 | 63.8269 |
| motorcycle | 24.6923 | 47.5495 | 33.8421 | 35.4143 |
| bicycle | 12.1445 | 33.2427 | 18.3169 | 18.1369 |
Links
Download results as .csv file
Benchmark page
