Cascaded Hierarchical Model

Cascaded hierarchical model is an image segmentation framework, which learns contextual information in a hierarchical framework. At each level of the hierarchy, a classifier is trained based on downsampled input images and outputs of previous levels. Our model then incorporates the resulting multi-resolution contextual information into a classifier to segment the input image at original resolution. We repeat this procedure by cascading the hierarchical framework to
improve the segmentation accuracy.

Matlab Codes

The source code and precompiled binaries for 64-bit linux are available here.
- If you want to compile the mex files use “compile.m” script available in the package.
- You should start with “TrainScript.m”.
- This code was written from scratch for the purpose of sharing.
- The results are not exactly the same as the reported numbers in the paper
(fortunately they are slightly better).

Edge Detection

CHM achieves near state-of-the-art performance on Berkeley dataset and outperforms state-of-the-art methods such as SE and SCG on NYU depth dataset v2.

The CHM code and corresponding scripts for edge detection for both BSDS500 and NYU depth datasets can be found here.

Precomputed testing results and evaluation files for BSDS 500 dataset.

Scene Labeling

On Stanford background dataset, CHM achieves 82.95% pixel accuracy.

The CHM code and corresponding scripts for scene labeling for the Stanford background dataset can be found here.

I applied the model trained on Stanford background dataset to a video recorded in state street, Salt Lake City.

If you use this package please cite the following:

M. Seyedhosseini, M. Sajjadi, and T. Tasdizen. Image segmentation with cascaded hierarchical models and logistic disjunctive normal networks. In ICCV 2013 (accepted). [pdf,bibtex]

- Our extended version of CHM paper got accepted to PAMI.
- I released the training and testing scripts for Scene labeling and edge detection. Feb 4th 2014
- An extended version of the ICCV paper with more results and theoretical insights can be found on arxiv. Feb 4th 2014 
- Version 2.0 was released (edge filters were added). December 29th 2013
- Version 1.0 was released. October 23rd 2013

If you have any questions please email me or Mehdi Sajjadi.

Precision-Recall curves for BSDS500 dataset.

Precision-Recall curves for NYU depth v2 dataset.

Test samples of scene labeling on the Stanford background dataset. First row: input image, Second row: CHM, Third row: CHM with intra-class connection, Fourth row: Groundtruth.