Skip to main content

Salient Object Detection: A Benchmark

Citations: 1,372 | Paper Published: 07 Oct 2015 | Institutions: University of Wisconsin–Milwaukee, University of Oxford, University of Massachusetts Amherst, Beihang University | Authors: Ali Borji1, Ming-Ming Cheng, Huaizu Jiang, Jia Li

Abstract in 2 Lines

The paper compares 41 state-of-the-art models for salient object detection and segmentation over seven challenging datasets to benchmark the performance of these methods. The evaluation shows significant progress in accuracy and running time over the last few years, and the paper proposes solutions for tackling open problems in the field.

Contributions​

  • Conducting a benchmarking study of 41 state-of-the-art models for salient object detection and segmentation over seven challenging datasets.
  • Analyzing the influences of center bias and scene complexity in model performance, and proposing probable solutions for tackling several open problems, such as evaluation scores and dataset bias.

Practical Implications​

  • The benchmarking study provides a comprehensive evaluation of state-of-the-art models for salient object detection and segmentation, which can guide researchers and practitioners in choosing the most suitable method for their specific application.
  • The analysis of the influences of center bias and scene complexity in model performance can help researchers better understand the limitations of current models and design more effective ones in the future.
  • The proposed solutions for tackling open problems, such as evaluation scores and dataset bias, can improve the reliability and generalizability of salient object detection and segmentation methods in real-world scenarios.

Methods​

Methodology
  • Benchmarking study: The paper conducts a benchmarking study of 41 state-of-the-art models for salient object detection and segmentation over seven challenging datasets. The models are evaluated based on various metrics, including precision, recall, F-measure, and mean absolute error.
  • Analysis of model performance: The paper analyzes the influences of center bias and scene complexity in model performance, and identifies hard cases for the state-of-the-art models.
  • Proposed solutions: The paper proposes probable solutions for tackling several open problems, such as evaluation scores and dataset bias, which can improve the reliability and generalizability of salient object detection and segmentation methods in real-world scenarios.

Data​

  • MSRA-B: A dataset of 5,000 images with pixel-level annotations.
  • MSRA-10K: A subset of MSRA-B with 10,000 images without annotations.
  • HKU-IS: A dataset of 4,447 images with pixel-level annotations.
  • DUT-OMRON: A dataset of 5,168 images with pixel-level annotations.
  • PASCAL-S: A subset of PASCAL VOC 2010 with 850 images and pixel-level annotations.
  • ECSSD: A dataset of 1,000 images with pixel-level annotations.
  • SOD: A dataset of 300 images with pixel-level annotations.

Results​

Performance Analysis
  • The top-performing model achieves an F-measure of 0.937 on the MSRA-B dataset.
  • The best model identified in the previous benchmark conducted three years ago achieved an F-measure of 0.882 on the same dataset.
  • The models designed specifically for salient object detection generally work better than models in closely related areas, such as objectness and fixation prediction.
  • The analysis of the influences of center bias and scene complexity in model performance shows that the models tend to have higher precision and recall for objects located near the center of the image and in simple scenes.

Limitations​

  • The paper only evaluates a limited number of state-of-the-art models, and there may be other models that perform better on these datasets.
  • The paper only evaluates the models on seven datasets, and there may be other datasets that are more challenging or representative of real-world scenarios.

Table of Contents