/VOS_Demo

Video Object Segmentation Demo

Video object segmentation is a task that given an input video, and then output each pixel's class(object class can be 1..N, background class is 0).

Visualization

Here, we show some visualization results(from DAVIS 2017[1]).

  • Input: a video;
  • Output: a mask set.
    • mask color represents pixel class.

1. Video: Camel

camel-img

camel-anno

2. Video: Soapbox

soapbox-img

soapbox-anno

Dataset

Evaluation on DAVIS 2017 val set[1]

DAVIS 2017 val set J mean J recall J decay F mean F recall F decay G mean
performance 0.6736 0.7989 0.1750 0.7273 0.8513 0.1981 0.7004
baseline[2] - - - - - - 0.6880
  • J means IoU;
  • F means F1-score;
  • More bigger mean and recall , the better;
  • Less decay, th better.

Reference

[1]Pont-Tuset J, Perazzi F, Caelles S, et al. The 2017 davis challenge on video object segmentation[J]. arXiv preprint arXiv:1704.00675, 2017. [2]Robinson A, Lawin F J, Danelljan M, et al. Learning Fast and Robust Target Models for Video Object Segmentation[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2020: 7406-7415.