This is a Matlab implementation of the layered object segmentation algorithm described in [1, 2]. The algorithm utilizes the object detection results obtained from deformable part-based models [3] with superpixels obtained from [4], and builds a Bayesian inference framework towards the semantic and instance segmentation of multi-class objects. The code is trained and tested using the PASCAL VOC dataset [5].

Acknowledgements: We graciously thank the authors of the previous code releases and image benchmarks for making them publically available.

The Matlab code is actually not runnable anymore. The repo is only used for a proof of concept.

The codes are mainly located in initialization, bias field and segmentation, where bias field contains the bias field learning code and segmentation contains the layered segmentation inference code.

[1] Y. Yang, S. Hallman, D. Ramanan, C. Fowlkes. Layered Object Detection for Multi-Class Segmentation. CVPR 2010.

[2] Y. Yang, S. Hallman, D. Ramanan, C. Fowlkes. Layered Object Models for Image Segmentation. PAMI 2012.

[3] P. Felzenszwalb, R. Girshick, D. McAllester, D. Ramanan. Discriminatively Trained Deformable Part Models. PAMI 2010.

[4] P. Arbelaez, M. Maire, C. Fowlkes, J. Malik. Contour Detection and Hierarchical Image Segmentation. PAMI 2011.

[5] M. Everingham, L. Van Gool, J. Winn, A. Zisserman. The PASCAL Visual Object Classes (VOC) Challenge. IJCV 2010.