Visual Explanation Method Combining Hierarchical Decision Information

doi:10.3778/j.issn.1002-8331.2209-0144

Abstract

Abstract: Visual explanation method is a hot research topic in the interpretable field of DNN, but the existing methods do not effectively combine hierarchical decision information, resulting in poor visual explanation effect. To solve this problem, this paper proposes a method combining hierarchical decision information. Firstly, the high-fine-grained local decision information is mined to generate weighted feature maps that are highly correlated with the decision results, and the sequential grouping method is used to merge them to obtain masks. Then the masks are processed by fuzzy boundary and integral method. And the importance scores based on global decision contribution are calculated in parallel, which improves the sensitivity of the algorithm to the global decision information and the speed of the algorithm. Finally, the optimal parameters are determined through ablation studies, and the method is qualitatively and quantitatively compared with the existing advanced methods on ImageNet dataset. The experimental results show that by combining hierarchical decision information, the proposed method achieves better results in the confidence test and location test, and it takes up to 68 ms.

Key words: visual explanation, hierarchical information, global decision contribution, class activation mapping (CAM), integral method

摘要： 视觉解释方法是深度神经网络可解释领域的热门研究课题，但现有方法未能有效利用多层次决策信息导致视觉解释效果差。针对此问题，提出一种融合多层次决策信息的视觉解释方法。挖掘特征图中高细粒度局部层次的决策信息生成一组与决策结果相关性强的加权特征图，采用定序分组方式对其合并，获取一组低冗余度掩码；采取模糊边界和积分方法对掩码进行处理，基于全局层次的决策贡献并行计算分组掩码重要度分数，提高了算法对全局决策信息的敏感性和算法速度。通过消融实验确定了算法的最优参数组合，并在ImageNet数据集上与现有的先进视觉解释方法进行了定性和定量比较。实验结果表明：该方法通过结合多层次决策信息，在置信度测试和定位测试中取得了更好的视觉解释结果，且耗时达到68 ms。

关键词: 视觉解释, 多层次信息, 全局决策贡献, 类激活映射（CAM）, 积分方法

YANG Chuandong, QIAN Lizhi, SUN Ziwen, CHEN Dong, LING Chong. Visual Explanation Method Combining Hierarchical Decision Information[J]. Computer Engineering and Applications, 2024, 60(3): 246-254.

杨传栋, 钱立志, 孙子文, 陈栋, 凌冲. 融合多层次决策信息的视觉解释方法[J]. 计算机工程与应用, 2024, 60(3): 246-254.

References

[1] 杨朋波, 桑基韬, 张彪, 等. 面向图像分类的深度模型可解释性研究综述[J]. 软件学报, 2023, 34（1）: 230-254.
YANG P B, SANG J T, ZHANG B, et al. Survey on interpretability of deep models for image classification[J]. Journal of Software, 2023, 34（1）: 230-254.
[2] ZHOU B, KHOSLA A, LAPEDRIZA A, et al. Learning deep features for discriminative localization[C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, 2016: 2921-2929.
[3] SELVARAJU R R, COGSWELL M, DAS A, et al. Grad-CAM: visual explanations from deep networks via gradient-based localization[C]//International Conference on Computer Vision, Piscataway, 2017: 618-626.
[4] SPRINGENBERG J T, DOSOVITSKIY A, BROX T, et al. Striving for simplicity: the all convolutional net[C]//International Conference on Learning Representations, San Diego, 2015: 1.
[5] CHATTOPADHYAY A, SARKAR A, HOWLADER P, et al. Grad-CAM++: improved visual explanations for deep convolutional networks[C]//Conference on Computer Vision and Pattern Recognition, Salt Lake City, 2018: 839-847.
[6] OMEIZA D, SPEAKMAN S, CINTAS C, et al. Smooth Grad-CAM++: an enhanced inference level visualization technique for deep convolutional neural network models[EB/OL]. （2019-08-03）. https://arxiv.org/pdf/1908.01224.pdf.
[7] WANG H, WANG Z, DU M, et al. Score-CAM: score-weighted visual explanations for convolutional neural networks[C]//Conference on Computer Vision and Pattern Recognition, Seattle, 2020: 24-25.
[8] NAIDU R, GHOSH A, MAURYA Y, et al. IS-CAM: integrated Score-CAM for axiomatic-based explanations[EB/OL]. （2020-10-06）. https://arxiv.org/pdf/2010.03023v1.pdf.
[9] LEE J R, KIM S, PARK I, et al. Relevance-CAM: your model already knows where to look[C]//Computer Vision and Pattern Recognition, 2021: 14944-14953.
[10] ZHANG Q, RAO L, YANG Y. Group-CAM: group score-weighted visual explanations for deep convolutional networks[EB/OL].（2021-06-19）. https://arxiv.org/pdf/2103. 13859.pdf.
[11] ZHENG Q, WANG Z, ZHOU J, et al. Shap-CAM: visual explanations for convolutional neural networks based on shapley value[C]//European Conference on Computer Vision, Tel Aviv, 2022: 459-474.
[12] 梁先明, 倪帆, 陈文洁, 等. 基于时频Grad-CAM的调制识别网络可解释研究[J/OL]. 西南交通大学学报: 1-9（2022-06-08）.https://kns.cnki.net/kcms/detail/51.1277.u.20220608.
1636.008.html.
LIANG X M, NI F, CHEN W J, et al. Interpretability of modulation recognition network based on time-frequency Grad-CAM[J/OL]. Journal of Southwest Jiaotong University: 1-9（2022-06-08）. https: //kns.cnki.net/kcms/detail/51.1277.u.20220608.1636.008.html.
[13] 张宇, 梁凤梅, 刘建霞. 融合类激活映射和视野注意力的皮肤病变分割[J]. 计算机工程与应用, 2023, 59（21）: 187-194.
ZHANG Y, LIANG F M, LIU J X. Skin lesion segmentation of based on classification activation mapping and visual field attention[J]. Computer Engineering and Applications, 2023, 59（21）: 187-194.
[14] RUSSAKOVSKY O, DENG J, SU H, et al. ImageNet large scale visual recognition challenge[J]. International Journal of Computer Vision, 2015, 115（3）: 211-252.
[15] ZEILER M D, FERGUS R. Visualizing and understanding convolutional networks[C]//European Conference on Computer Vision. Switzerland: Springer, 2014: 818-833.
[16] YANG L, JIANG H, CAI R, et al. CondenseNet V2: sparse feature reactivation for deep networks[EB/OL].[2021-04-09]. https: //arxiv.org/pdf/2104.04382. pdf.
[17] SUNDARARAJAN M, TALY A, YAN Q. Axiomatic attribution for deep networks[C]//International Conference on Machine Learning, Sydney, 2017: 3319-3328.
[18] STURMFELS P, LUNDBERG S, LEE S I. Visualizing the impact of feature attribution baselines[J]. Distill, 2020, 5（1）: 1.
[19] SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[C]//International Conference on Learning Representations, San Diego, 2015: 4.
[20] POPPI S, CORNIA M, BARALDI L, et al. Revisiting the evaluation of class activation mapping for explainability: a novel metric and experimental analysis[C]//Conference on Computer Vision and Pattern Recognition, 2021: 2299-2304.
[21] JIANG P T, ZHANG C B, HOU Q, et al. LayerCAM: exploring hierarchical class activation maps for localization[J]. IEEE Transactions on Image Processing, 2021, 30: 5875-5888.
[22] PETSIUK V, DAS A, SAENKO K. RISE: randomized input sampling for explanation of black-box models[C]//British Machine Vision Conference, Newcastle, 2018: 151.
[23] NIE W, YANG Z, PATEL A. A theoretical explanation for perplexing behaviors of backpropagation-based visualizations[C]//International Conference on Machine Learning, Stockholm, 2018: 3809-3818.