PySODMetrics: A simple and efficient implementation of SOD metrics

Introduction

A simple and efficient implementation of SOD metrics.

• Based on numpy and scipy
• Verification based on Fan's matlab code https://github.com/DengPingFan/CODToolbox
• The code structure is simple and easy to extend
• The code is lightweight and fast

Your improvements and suggestions are welcome.

Related Projects

• PySODEvalToolkit: A Python-based Evaluation Toolbox for Salient Object Detection and Camouflaged Object Detection

Supported Metrics

| Metric | Sample-based | Whole-based | Related Class | | ----------------------------------------- | ------------------------------------------- | ------------------------ | ------------------------------------- | | MAE | soft,si-soft | | MAE | | S-measure $S{m}$ | soft | | Smeasure | | weighted F-measure ($F^{\omega}{\beta}$) | soft | | WeightedFmeasure | | Human Correction Effort Measure | soft | | HumanCorrectionEffortMeasure | | Multi-Scale IoU | max,avg,adp,bin | | MSIoU | | E-measure ($E{m}$) | max,avg,adp | | Emeasure | | F-measure (old) ($F{beta}$) | max,avg,adp | | Fmeasure (Will be removed!) | | F-measure (new) ($F{beta}$, $F{1}$) | max,avg,adp,bin,si-max,si-avg,si-adp,si-bin | bin,si-max,si-avg,si-bin | FmeasureV2+FmeasureHandler | | BER | max,avg,adp,bin,si-max,si-avg,si-adp,si-bin | bin,si-max,si-avg,si-bin | FmeasureV2+BERHandler | | Dice | max,avg,adp,bin,si-max,si-avg,si-adp,si-bin | bin,si-max,si-avg,si-bin | FmeasureV2+DICEHandler | | FPR | max,avg,adp,bin,si-max,si-avg,si-adp,si-bin | bin,si-max,si-avg,si-bin | FmeasureV2+FPRHandler | | IoU | max,avg,adp,bin,si-max,si-avg,si-adp,si-bin | bin,si-max,si-avg,si-bin | FmeasureV2+IOUHandler | | Kappa | max,avg,adp,bin,si-max,si-avg,si-adp,si-bin | bin,si-max,si-avg,si-bin | FmeasureV2+KappaHandler | | Overall Accuracy | max,avg,adp,bin,si-max,si-avg,si-adp,si-bin | bin,si-max,si-avg,si-bin | FmeasureV2+OverallAccuracyHandler | | Precision | max,avg,adp,bin,si-max,si-avg,si-adp,si-bin | bin,si-max,si-avg,si-bin | FmeasureV2+PrecisionHandler | | Recall | max,avg,adp,bin,si-max,si-avg,si-adp,si-bin | bin,si-max,si-avg,si-bin | FmeasureV2+RecallHandler | | Sensitivity | max,avg,adp,bin,si-max,si-avg,si-adp,si-bin | bin,si-max,si-avg,si-bin | FmeasureV2+SensitivityHandler | | Specificity | max,avg,adp,bin,si-max,si-avg,si-adp,si-bin | bin,si-max,si-avg,si-bin | FmeasureV2+SpecificityHandler | | TNR | max,avg,adp,bin,si-max,si-avg,si-adp,si-bin | bin,si-max,si-avg,si-bin | FmeasureV2+TNRHandler | | TPR | max,avg,adp,bin,si-max,si-avg,si-adp,si-bin | bin,si-max,si-avg,si-bin | FmeasureV2+TPRHandler |

NOTE: - Sample-based si- variants calculate the sample-specific mean/maximum based on the sample-based fm sequence with a shape of (num_targets, 256). - Whole-based si- variants calculate the mean/maximum based on the average fm sequence across all targets from all samples. - Because the *adp variants are specialized for sample-based computation, they do not support whole-based computation.

Usage

The core files are in the folder py_sod_metrics.

• [Latest, but may be unstable] Install from the source code: pip install git+https://github.com/lartpang/PySODMetrics.git
• [More stable] Install from PyPI: pip install pysodmetrics

Examples

• examples/test_metrics.py
• examples/metric_recorder.py

Reference

• Matlab Code by DengPingFan(https://github.com/DengPingFan): In our comparison (the test code can be seen under the test folder), the result is consistent with the code.
  • The matlab code needs to change Bi_sal(sal>threshold)=1; to Bi_sal(sal>=threshold)=1; in https://github.com/DengPingFan/CODToolbox/blob/910358910c7824a4237b0ea689ac9d19d1958d11/Onekey_Evaluation_Code/OnekeyEvaluationCode/main.m#L102. For related discussion, please see the issue.
  • 2021-12-20 (version 1.3.0): Due to the difference between numpy and matlab, in version 1.2.x, there are very slight differences on some metrics between the results of the matlab code and ours. The recent PR alleviated this problem. However, there are still very small differences on E-measure. The results in most papers are rounded off to three or four significant figures, so, there is no obvious difference between the new version and the version 1.2.x for them.
• https://en.wikipedia.org/wiki/Precision_and_recall

```text @inproceedings{Fmeasure, title={Frequency-tuned salient region detection}, author={Achanta, Radhakrishna and Hemami, Sheila and Estrada, Francisco and S{\"u}sstrunk, Sabine}, booktitle=CVPR, number={CONF}, pages={1597--1604}, year={2009} }

@inproceedings{MAE, title={Saliency filters: Contrast based filtering for salient region detection}, author={Perazzi, Federico and Kr{\"a}henb{\"u}hl, Philipp and Pritch, Yael and Hornung, Alexander}, booktitle=CVPR, pages={733--740}, year={2012} }

@inproceedings{Smeasure, title={Structure-measure: A new way to evaluate foreground maps}, author={Fan, Deng-Ping and Cheng, Ming-Ming and Liu, Yun and Li, Tao and Borji, Ali}, booktitle=ICCV, pages={4548--4557}, year={2017} }

@inproceedings{Emeasure, title="Enhanced-alignment Measure for Binary Foreground Map Evaluation", author="Deng-Ping {Fan} and Cheng {Gong} and Yang {Cao} and Bo {Ren} and Ming-Ming {Cheng} and Ali {Borji}", booktitle=IJCAI, pages="698--704", year={2018} }

@inproceedings{wFmeasure, title={How to evaluate foreground maps?}, author={Margolin, Ran and Zelnik-Manor, Lihi and Tal, Ayellet}, booktitle=CVPR, pages={248--255}, year={2014} }

@inproceedings{MSIoU, title = {Multiscale IOU: A Metric for Evaluation of Salient Object Detection with Fine Structures}, author = {Ahmadzadeh, Azim and Kempton, Dustin J. and Chen, Yang and Angryk, Rafal A.}, booktitle = ICIP, year = {2021}, }

@inproceedings{SizeInvarianceVariants, title = {Size-invariance Matters: Rethinking Metrics and Losses for Imbalanced Multi-object Salient Object Detection}, author = {Feiran Li and Qianqian Xu and Shilong Bao and Zhiyong Yang and Runmin Cong and Xiaochun Cao and Qingming Huang}, booktitle = ICML, year = {2024} }

@inproceedings{HumanCorrectionEffortMeasure, title = {Highly Accurate Dichotomous Image Segmentation}, author = {Xuebin Qin and Hang Dai and Xiaobin Hu and Deng-Ping Fan and Ling Shao and Luc Van Gool}, booktitle = ECCV, year = {2022} } ```