Method

Improving Semantic Segmentation via Video Propagation and Label Relaxation [VideoProp-LabelRelax]


Submitted on 17 Jun. 2019 20:04 by
Karan Sapra (NVIDIA)

Running time:n s
Environment:GPU @ 1.5 Ghz (Python + C/C++)

Method Description:


In this paper, we present a video prediction-
based methodology to scale up training sets by
synthesizing new training samples in order to
improve the accuracy of semantic segmentation
networks. We exploit video prediction models'
ability to predict future frames in order to
also predict future labels. A joint propagation
strategy is also proposed to alleviate mis-
alignments in synthesized samples. We
demonstrate that training segmentation models on
datasets augmented by the synthesized samples
leads to significant improvements in accuracy.
Furthermore, we introduce a novel boundary label
relaxation technique that makes training robust
to annotation noise and propagation artifacts
along object boundaries. Our proposed methods
achieve state-of-the-art mIoUs of 83.5% on
Cityscapes and 82.9% on CamVid. Our single
model, without model ensembles, achieves 72.8%
mIoU on the KITTI semantic segmentation test
set, which surpasses the winning entry of the
ROB challenge 2018.
Parameters:

Details can be found here:
Arxiv Link: https://arxiv.org/abs/1812.01593
Latex Bibtex:
@inproceedings{semantic_cvpr19,
author = {Yi Zhu*, Karan Sapra*, Fitsum
A. Reda, Kevin J. Shih, Shawn Newsam, Andrew
Tao, Bryan Catanzaro},
title = {Improving Semantic
Segmentation via Video Propagation and Label
Relaxation},
booktitle = {IEEE Conference on Computer
Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2019},
url = {https://nv-
adlr.github.io/publication/2018-Segmentation}
}

Detailed Results

This page provides detailed results for the method(s) selected. For the first 20 test images, we display the original image, the color-coded result and an error image. The error image contains 4 colors:
red: the pixel has the wrong label and the wrong category
yellow: the pixel has the wrong label but the correct category
green: the pixel has the correct label
black: the groundtruth label is not used for evaluation

Test Set Average

IoU class iIoU class IoU category iIoU category
72.82 48.68 88.99 75.26
This table as LaTeX

Test Image 0

Input Image

Prediction

Error


Test Image 1

Input Image

Prediction

Error


Test Image 2

Input Image

Prediction

Error


Test Image 3

Input Image

Prediction

Error


Test Image 4

Input Image

Prediction

Error


Test Image 5

Input Image

Prediction

Error


Test Image 6

Input Image

Prediction

Error


Test Image 7

Input Image

Prediction

Error


Test Image 8

Input Image

Prediction

Error


Test Image 9

Input Image

Prediction

Error




eXTReMe Tracker