The Lovász-Softmax loss: A tractable surrogate for the optimization of the intersection-over-union measure in neural networks

Benchmark Model Rank Results
real-time-semantic-segmentation-on-cityscapesENet + Lovász-Softmax#36mIoU: 63.1%Frame (fps): 76.9Time (ms): 13
semantic-segmentation-on-cityscapesENet + Lovász-Softmax#91Mean IoU (class): 63.06%
semantic-segmentation-on-pascal-voc-2012Deeplab-v2 with Lovasz-Softmax loss#32Mean IoU: 79.00%
semantic-segmentation-on-pascal-voc-2012Deeplab-v2 + Lovász-Softmax#33Mean IoU: 79.0%