Home

Awesome

TIS: Tukey-Inspired Video Object Segmentation

Contact: Brent Griffin (griffb at umich dot edu)

Publication

Tukey-Inspired Video Object Segmentation<br /> Brent A. Griffin and Jason J. Corso<br /> IEEE Winter Conference on Applications of Computer Vision (WACV), 2019

Please cite our paper if you find it useful for your research.

@inproceedings{GrCoWACV2019,
  author = {Griffin, Brent A. and Corso, Jason J.},
  booktitle = {IEEE Winter Conference on Applications of Computer Vision (WACV)},
  title = {Tukey-Inspired Video Object Segmentation},
  year = {2019}
}

Method

Video Description: https://youtu.be/FeWnFz4Cf_8

IMAGE ALT TEXT HERE

TIS processes image data to find foreground objects. The outlier scale acts as a weighting that adapts to frame-to-frame video characteristics. In this example, we focus on optical flow magnitude with outliers depicted as black pixels (middle row). Flow distributions are offset from the median (bottom row) and include the interquartile range (solid lines) and outlier thresholds (dotted lines). alt text <br />

TIS_M processes and combines multiple segmentation masks, generating a collectively more robust method of segmentation. alt text

Results

DAVIS results for state-of-the-art unsupervised methods. TIS-based methods achieve top results in all categories. alt text <br />

Visual comparison of segmentation methods on DAVIS dataset. TIS_M-based methods improve performance across all categories of supervision. alt text

Pre-Computed Results

Pre-computed results for TIS_0, TIS_S, and TIS_M on DAVIS 2016 are provided in the /precomputed_results folder.

Source Code

Source code for TIS_0 and TIS_S segmentation methods from the paper is provided in the /TIS folder.

Source code for the TIS_M segmentation method from the paper is provided in the /TIS_M folder.

Use

This code is available for non-commercial research purposes only.