Quantum-Cuts: State-Of-The-Art Automatic Saliency Map Generation and Sailent Object Segmentation



Quantum-Cuts (QCUT) is a recent saliency map generation technique. It automatically detects the most interesting (e.g. the object) region in an image and can highlight where people tend to look at first or at the most. It is inspired from and based on Quantum Mechanics. QCUT provides a parameter-free -hence dataset independent-, unsupervised and fully automatic saliency map generation, which outperforms current state-of-the-art algorithms. Its salient object segmentation results exhibit such a promising accuracy that pushes the frontier in this field to the borders of the input-driven processing only – without the use of “object knowledge” aided by long-term human memory and intelligence. Furthermore, with the near-future technologies for measuring a quantum wavefunction, QCUT may have a unique potential: Automatic object segmentation in an actual physical setup in nano-scale. Such an unprecedendent property would not only produce segmentation results instantaneously, but may be a unique opportunity to achieve accurate object segmentation in real-time for the massive visual repositories of today’s “Big Data”.

QCUT provides the following advantages over existing algorithms:
1. It performs repeatedly the best in all publicly available datasets over all existing saliency detection algorithms.
2. It is quite fast (around 0.4 seconds for an average image of size 400*300)
3. It can propose more than one region. Coverage of 5 proposals can achieve significant performance upgrade.
4. It guarantees a certain accuracy of background. Thus, it is available for post-processing algorithms such as Grab-Cut for further upgrade where you can define an exact background.
5. Coverage of 5 saliency maps with Grab-Cut post-processing can achieve dominating performance (%20 better precision at same recall, %30 better recall at same precision on average than the 2nd best algorithm)

Live Quantum-Cuts Demo

Try QCUT yourself! You have three options:
1) Drag & drop any image to the box below.
2) Copy and Paste any image URL (from any website).
3) (Batch Processing) Drag & drop a zip file containing image(s).
For the options 1 and 2, wait for few seconds to see the TOP-1 (upper row) and TOP-5 (all) results of the saliency and object maps. For the option 3, another zip file including the TOP-10 results file will be produced and you can download it after a while (Warning: If the Chrome gets into a timeout state, simply press “Wait” to retrieve the results – Do NOT select “Kill pages”, otherwise you will lose them!).
1) You can use the slider bar to change the threshold that will be used to generate object map(s) from the saliency map(s). For option 3, you need to set it before you submit the zip file.
2) Please do not add more than 100 images in the zip file.

Paste Image URL:

Drag an image into the box below

Saliency Map Output 1

Object Binary Map 1

Saliency Map 2

Object Binary Map 2

Saliency Map 3

Object Binary Map 3

Saliency Map 4

Object Binary Map 4

Saliency Map 5

Object Binary Map 5


Please wait....

Contact QCUT Developers:
Çağlar Aytekin: caglar -dot - aytekin -AT- tut -dot- fi
Prof. Serkan Kiranyaz: serkan - dot- kiranyaz -AT- tut -dot- fi
Prof. Moncef Gabbouj: moncef - dot - gabbouj -AT- tut -dot- fi

QCut SW Team :
Ezgi Can Ozan: ezgi -dot - ozan --AT- tut -dot - fi
Dr. Iftikhar Ahmad: iftikhar - dot- ahmad --AT -- tut - dot- fi

Quantitative Performance Comparison of Quantum-Cuts Against the State-Of-The-Art

Below: Below: Precision-recall curves of the best performing state-of-the art algorithms: GBM, MARKOV and Quantum-Cuts (QCUT) on various databases..


MSRA: 1000 photos with pixel-wise ground truth.
R. Achanta, S. Hemami, F. Estrada and S. Süsstrunk, Frequency-tuned Salient Region Detection, CVPR 2009.
MSRA5000: 5000 photos with bounding box ground-truth.
Tie Liu, Jian Sun, Nan-Ning Zheng, Xiaoou Tang and Heung-Yeung Shum. Learning to Detect A Salient Object, CVPR 2007.
PASCAL: 1500 photos with pixel-wise ground truth.
W.B. Zou, K. Kpalma, Z. Liu and J. Ronsin. Segmentation Driven Low-rank Matrix Recovery for Saliency Detection, BMVC 2013
ECSSD: 1000 photos with pixel-wise ground truth.
Qiong Yan, Li Xu, Jianping Shi, Jiaya Jia. Hierarchical Saliency Detection. CVPR 2013.
DUTUMRON: 5168 photos with pixel-wise ground truth.
C. Yang, L. Zhang, H. Lu, X. Ruan and M-H. Yang, Saliency Detection via Graph-Based Manifold Ranking, CVPR, 2013.
BSD: 300 photos with pixel-wise ground-truth.
V. Movahedi and J. H. Elder, Design and Perceptual Validation of Performance Measures for Salient Object Segmentation, POCV 2010.

Methods in the Plot:
QCUT TOP 1: Quantum-Cuts, QCUT TOP 1-PP: Quantum-Cuts after post-processing, QCUT TOP 5-PP: Best result from 5 proposals obtained by Quantum-Cuts after post-processing
GBM: C. Yang, L. Zhang, H. Lu, X. Ruan and M-H. Yang, “Saliency Detec-tion via Graph-Based Manifold Ranking,” CVPR, 2013.
MARKOV: B. Jiang, L. Zhang, H. Lu, C. Yang and M-H. Yang, “Saliency Detec-tion via Absorbing Markov Chain,” 14th International Conference on Computer Vision, 2013.

Quantitative Performance Comparison of Quantum-Cuts Against the State-Of-The-Art (GRAND DATASET)

GRAND DATASET: Combination of all 6 datasets, containing 13968 images.

Visual Performance Comparison Against The Current State-Of-The-Art

Visual comparison of saliency maps generated by the Quantum Cuts and the two best competitors, GBM and MARKOV.

On Videos: Where is the salient object?


Ground Truth

QCUT Output

More Videos

On NEWSPAPERS: Where does a Reader first look?

The Salient Region  
The Salient Region  

More examples

On Social Networks: Where do People first look?

The Salient Region 
Object Extracted  

More examples

Daily Photos: Where is the object?

Saliency Map  
Object Extracted  

More daily photos