Comparison of different methods


In this supplementary material, we provide ground truth and results of our algorithm and eight state-of-the-art saliency algorithms on our evaluation dataset. Due to the limitation in the size of the supplementary material, we resize all the images so that the maximum dimension per image is 200 pixels. Note that for the results reported in the paper, we use the images in their original size when computing the performance of each method.


The eight state-of-the-art saliency algorithms are:
AMC: B. Jiang, L. Zhang, H. Lu, M.H. Yang, C. Yang: Saliency detection via absorbing markov chain. In ICCV, (2013)
CH: X. Li, Y. Li, C. Shen, A. Dick, A. van den Hengel: Contextual hypergraph modeling for salient object detection. In ICCV, (2013)
GBMR: C. Yang, L. Zhang, H. Lu, X. Ruan, M.H. Yang: Saliency detection via graph-based manifold ranking. In CVPR, (2013)
SMVJ: M. Cerf, J. Harel, W. Einhauser, C. Koch: Predicting human gaze using low-level saliency combined with face detection. In NIPS, (2007)
LR: X. Shen, Y. Wu: A unified approach to salient object detection via low rank matrix recovery. In CVPR, (2012)
Judd: T. Judd, K. Ehinger, F. Durand, A. Torralba: Learning to predict where humans look. In ICCV, (2009)
Borji: A. Borji: Boosting bottom-up and top-down visual features for saliency estimation. In CVPR, (2012)
SC: S. Karthikeyan, V. Jagadeesh, B.S. Manjunath: Learning top down scene context for visual attention modelling in natural images. In ICIP, (2013)


Click on one of the links below to see the results of all nine methods as well as the ground truth. The last link illustrates some failure cases.

Images 1 -100
Images 101 -200
Images 201 -300
Images 301 -400
Images 401 -500
Images 501 -600
Images 601 -700
Images 701 -800
Failed samples