Method Details

Details for method 'Semantic Instance Segmentation with a Discriminative Loss Function'


Method overview

name Semantic Instance Segmentation with a Discriminative Loss Function
challenge instance-level semantic labeling
details This method uses a discriminative loss function, operating at the pixel level, that encourages a convolutional network to produce a representation of the image that can easily be clustered into instances with a simple post-processing step. The loss function encourages the network to map each pixel to a point in feature space so that pixels belonging to the same instance lie close together while different instances are separated by a wide margin. Previously listed as "PPLoss".
publication Semantic Instance Segmentation with a Discriminative Loss Function
Bert De Brabandere, Davy Neven, Luc Van Gool
Deep Learning for Robotic Vision, workshop at CVPR 2017
project page / code
used Cityscapes data fine annotations
used external data ImageNet
runtime n/a
subsampling 2
submission date March, 2017
previous submissions


Average results

Metric Value
AP 17.4741
AP50% 35.8662
AP100m 27.8474
AP50m 30.9631


Class results

Class AP AP50% AP100m AP50m
person 13.4739 31.9628 25.0922 25.1403
rider 16.1671 40.7139 27.4539 28.1786
car 24.4367 43.1986 39.9743 44.0233
truck 16.7724 28.5141 24.3667 28.5748
bus 23.8738 39.1294 39.3879 47.7435
train 19.1593 35.6574 26.475 32.4971
motorcycle 15.2178 37.9132 22.1758 23.5113
bicycle 10.6918 29.8407 17.8536 18.0359



Download results as .csv file

Benchmark page