The Secrets of Salient Object Segmentation

By Yin Li, Xiaodi Hou, Christof Koch, James M. Rehg, Alan L. Yuille. 2014 IEEE Conference on Computer Vision and Pattern Recognition, 2014

1,322 citations recorded by the source corpus.

In this paper we provide an extensive evaluation of fixation prediction and salient object segmentation algorithms as well as statistics of major datasets. Our analysis identifies serious design flaws of existing salient object benchmarks, called the dataset design bias, by over emphasising the stereotypical concepts of saliency. The dataset design bias does not only create the discomforting disconnection between fixations and salient object segmentation, but also misleads the algorithm designing. Based on our analysis, we propose a new high quality dataset that offers both fixation and salient object segmentation ground-truth. With fixations and salient object being presented simultaneously, we are able to bridge the gap between fixations and salient objects, and propose a novel method for salient object segmentation. Finally, we report significant benchmark progress on 3 existing datasets of segmenting salient objects.

Subfields: Cognitive Neuroscience, Computer Vision and Pattern Recognition, Sensory Systems

View Paper: https://www.semanticscholar.org/paper/40e71c10edc2afc68d079546e8f4952cd52dc671

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Cognitive Neuroscience

1. Huchuan Lu
2. Ming-Ming Cheng
3. Deng-Ping Fan
4. Qingming Huang
5. Jianbing Shen
6. Ali Borji
7. Wenguan Wang
8. Zhe Wu

In Computer Vision and Pattern Recognition

1. Luc Van Gool
2. Ming-Hsuan Yang
3. Ming-Ming Cheng
4. Ling Shao
5. Dacheng Tao
6. Huchuan Lu
7. Xiaogang Wang
8. Yi Yang

In Sensory Systems

1. Jianbing Shen
2. Wenguan Wang
3. Huchuan Lu
4. Ming-Ming Cheng
5. Junwei Han
6. Deng-Ping Fan
7. Ali Borji
8. G. Wang

Citations in the corpus

• SALICON: Reducing the Semantic Gap in Saliency Prediction by Adapting Deep Neural Networks (2015, methodology): “We compare our method with 14 saliency prediction models , which are BMS [42], Adaptive Whitening Saliency (AWS) [11], Attention based on Information Maximization (AIM) [2], RARE [30], Local and Global saliency (LG) [1], eDN [39], Judd's model [21], image Signature Saliency (SigSal) [15], Context-Aw”
• Deep Level Sets for Salient Object Detection (2017, background): “The PASCAL [32] has 850 images with complex scene.”
• Shifting More Attention to Video Salient Object Detection (2019, background): “Previous studies [6, 45] quantitatively confirmed that there exists a strong correlation between such explicit, object-level saliency judgment (object-saliency) and the implicit visual attention allocation behavior (visual attention mechanism).”
• Rethinking RGB-D Salient Object Detection: Models, Data Sets, and Large-Scale Benchmarks (2019, background): “Center Bias: Center bias has been identified as one of the most significant biases of saliency detection datasets [91].”
• Non-local Deep Features for Salient Object Detection (2017, methodology): “08 s, a speed gain of 18 to 100 times as compared to other state-of-the-art deep learning methods, while being on par with state-of-the-art evaluation performance on the MSRA-B[30], HKU-IS[25], PASCAL-S[27], DUT-OMRON[49], ECSSD [48] and SOD [32] benchmark datasets.”
• Saliency detection by multi-context deep learning (2015, methodology): “A large number of approaches [63, 52, 40, 39, 32, 35, 60, 57, 56, 47, 41, 31, 27, 25, 24, 23, 11, 44, 17, 8, 13, 1, 21] are proposed to capture different saliency cues.”
• New Contour Cue-Based Hybrid Sparse Learning for Salient Object Detection (2019, methodology): “To achieve comprehensive evaluation, we conduct extensive experiments on four benchmark saliency datasets, including PASCAL-S (850 images) [64], ECSSD (1000 images) [65], DUT-OMRON (5168 images) [30], and THUS-10000 (10 000 images) [17].”
• A Simple Pooling-Based Design for Real-Time Salient Object Detection (2019, methodology): “As can be seen, using the boundaries of salient objects as supervision results in no improvement while using standard boundaries can greatly boost the performance on all three PASCAL-S [21] DUT-O [42] SOD [30]”
• Zoom Better to See Clearer: Human and Object Parsing with Hierarchical Auto-Zoom Net (2015, methodology): “This motivates us to work with PASCAL images [10] because these were chosen for studying multiple visual tasks, do not suffer from dataset design bias [18], and include large variations of objects, particularly of scale.”
• Dense and Low-Rank Gaussian CRFs Using Deep Embeddings (2017, methodology): “The evaluation criterion is the maximal F-Measure as in [19, 23].”
• Amulet: Aggregating Multi-level Convolutional Features for Salient Object Detection (2017, background): “PASCAL-S [26].”
• Salient Object Detection: A Discriminative Regional Feature Integration Approach (2013, methodology): “[62] propose to generate a saliency map by adaptively averaging the object proposals [63] with their foreground probabilities that are learned based on eye fixations features using the Random Forest regressor.”
• Enhanced-alignment Measure for Binary Foreground Map Evaluation (2018, methodology): “The employed datasets include PASCAL-S [Li et al. , 2014], ECSSD [Xie et al. , 2013], HKU-IS [Li and Yu, 2015], and SOD [Martin et al. , 2001].”
• Understanding Deep Learning Techniques for Image Segmentation (2019, background): “nopathy Image Dataset [169] ISLES: Ischemic Stroke Lesion Segmentation [135] Saliency MSRA Salient Object Database [37] Detection ECSSD: Extended Complex Scene Saliency Dataset [187] PASCAL-S DATASET [117] THUR15K: Group Saliency in Image [36] JuddDB: MIT saliency benchmark [18] DUT-OMRON Image Data”
• Interactive Two-Stream Decoder for Accurate and Fast Saliency Detection (2020, methodology): “Meanwhile, others datasets, like SOD [25], PASCAL-S [19], ECSSD [40], HKU-IS [17] and DUT-O [41], are only employed as test sets due to their relatively small scales, which contain 300, 850, 1000, 4447 and 5168 images respectively.”
• Deep Visual Attention Prediction (2017, methodology): “We evaluate the contribution of each component of the proposed deep attention model on TORONTO [25] and PASCAL-S [26] datasets and measure the performance using the shuffled AUC metric.”
• Visual attention network (2022, methodology): “We replace the backbone with VAN and hold experiments on common saliency detection benchmarks, including DUTS [126], DUT-O [127], and PASCALS [128].”