Explainable Computer Vision: Interpretability Methods for Deep Neural Networks Using Grad-CAM, SHAP, LIME, and Attention Visualization

References

[1] R. R. Selvaraju, M. Cogswell, A. Das, R. Vedantam, D. Parikh, and D. Batra, "Grad-CAM: Visual explanations from deep networks via gradient-based localization," in Proc. IEEE/CVF Int. Conf. Comput. Vis. (ICCV), Venice, Italy, Oct. 2017, pp. 618–626.

[2] T. Miller, "Explanation in artificial intelligence: Insights from the social sciences," J. Artif. Intell. Res., vol. 52, pp. 1–66, Jun. 2021.

[3] K. He, X. Zhang, S. Ren, and J. Sun, "Deep residual learning for image recognition," in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), Las Vegas, NV, USA, Jun. 2016, pp. 770–778.

[4] S. Ghassemi, A. Oakden-Rayner, and A. Beam, "The false hope of explainability in medical AI," Nat. Med., vol. 27, pp. 207–209, Feb. 2021.

[5] B. Buolamwini and T. Gebru, "Gender shades: Intersectional accuracy disparities in commercial gender classification," in Conf. Fairness Accountability Transparency, New York, NY, USA, Feb. 2018, pp. 77–91.

[6] M. D. Zeiler and R. Fergus, "Visualizing and understanding convolutional networks," in Proc. Eur. Conf. Comput. Vis. (ECCV), Zurich, Switzerland, Sep. 2014, pp. 818–833.

[7] A. Dosovitskiy et al., "An image is worth 16×16 words: Transformers for image recognition at scale," in Proc. Int. Conf. Learn. Represent. (ICLR), Virtual, May 2021.

[8] S. Lundberg and S. Lee, "A unified approach to interpreting model predictions," in Proc. 31st Conf. Neural Inf. Process. Syst. (NeurIPS), Long Beach, CA, USA, Dec. 2017, pp. 4768–4777.

[9] S. Shrikumar, A. Greenfield, P. Y. Kandel, and K. Q. Weinberger, "Computationally efficient measures of internal neuron importance," in Proc. Int. Conf. Mach. Learn. (ICML), Long Beach, CA, USA, Jun. 2019.

[10] I. E. Kumar, Y. Su, and B. Guestrin, "Shapley explanation of black-box models through example-based approximation," in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), Nashville, TN, USA, Jun. 2021, pp. 8611–8620.

[11] M. T. Ribeiro, S. Singh, and C. Guestrin, "'Why should I trust you?': Explaining the predictions of any classifier," in Proc. 22nd ACM SIGKDD Int. Conf. Knowl. Discovery Data Mining, San Francisco, CA, USA, Aug. 2016, pp. 1135–1144.

[12] A. Simonyan, A. Vedaldi, and A. Zisserman, "Deep inside convolutional networks: Visualising image classification models and saliency maps," in Int. Conf. Learn. Represent. (ICLR), Banff, AB, Canada, Apr. 2014, pp. 1–8.

[13] Y. Li, L. Su, and K. Li, "Context-aware attention mechanisms for computer vision," IEEE Trans. Pattern Anal. Mach. Intell., vol. 43, no. 6, pp. 1884–1903, Jun. 2021.

[14] A. Vaswani et al., "Attention is all you need," in Proc. 31st Conf. Neural Inf. Process. Syst. (NeurIPS), Long Beach, CA, USA, Dec. 2017, pp. 5998–6008.

[15] J. Clark, U. Khandelwal, O. Levy, and C. D. Manning, "What does BERT look at? An analysis of BERT's attention," arXiv preprint arXiv:1906.04341, Jun. 2019.

[16] T. Abnar and W. Zuidema, "Quantifying attention flow in transformers," in Proc. 58th Annu. Meeting Assoc. Comput. Linguistics, Seattle, WA, USA, Jul. 2020, pp. 4190–4200.

[17] J. Peng et al., "Comparative evaluation of image saliency models," in Proc. Eur. Conf. Comput. Vis. (ECCV), Munich, Germany, Sep. 2018, pp. 19–34.

[18] Y. Nie, A. Jiang, and Z. Li, "Perturbation-based explanation in machine learning systems," IEEE Trans. Knowl. Data Eng., vol. 32, no. 11, pp. 2223–2237, Nov. 2020.

[19] O. Russakovsky, J. Deng, H. Su, J. Krause, S. Satheesh, S. Ma, Z. Huang, A. Karpathy, A. Khosla, M. Bernstein, et al., "ImageNet large scale visual recognition challenge," Int. J. Comput. Vis., vol. 115, no. 3, pp. 211–252, Dec. 2015.

[20] U. Kaur, B. Singh, and N. Batra, "Explaining and improving adversarial robustness in image classification models," in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), Nashville, TN, USA, Jun. 2021, pp. 8211–8220.

[21] S. A. Seshia, D. Sadigh, and S. S. Sastry, "Towards verified artificial intelligence," in Int. J. Softw. Tools Technol. Transf., vol. 22, no. 3, pp. 325–344, Jun. 2020.

[22] M. D. Zeiler and R. Fergus, "Visualizing and understanding convolutional networks," in Proc. Eur. Conf. Comput. Vis. (ECCV), Zurich, Switzerland, Sep. 2014, pp. 818–833.

[23] A. E. Mahfouz, W. Du, and A. Malinovskiy, "Evaluating the robustness of convolutional neural networks for image classification under adversarial perturbations," IEEE Trans. Neural Netw. Learn. Syst., vol. 31, no. 8, pp. 2976–2987, Aug. 2020.

[24] J. Heo, S. Joo, and T. Moon, "Fooling neural network interpretations," in Proc. 25th ACM SIGKDD Int. Conf. Knowl. Discovery Data Mining, Anchorage, AK, USA, Aug. 2019, pp. 2280–2290.

[25] F. Doshi-Velez and B. Kim, "Towards a rigorous science of interpretable machine learning," arXiv preprint arXiv:1702.08608, Feb. 2017.

@article{1719860,
      author = {Tanya Verma, Lipakshi, Shivam Sharma, Nishtha Singh, Jayanti},
      title = {Explainable Computer Vision: Interpretability Methods for Deep Neural Networks Using Grad-CAM, SHAP, LIME, and Attention Visualization},
      journal = {Iconic Research And Engineering Journals},
      year = {2023},
      volume = {7},
      number = {3},
      pages = {894-900},
      issn = {2456-8880},
      url = {https://www.irejournals.com/formatedpaper/1719860.pdf},
      abstract = {The remarkable success of deep neural networks in computer vision tasks has been accompanied by a critical challenge: understanding why these "black box" models make specific predictions. This paper presents a comprehensive review of explainability methods for computer vision systems, focusing on four major interpretation techniques: Gradient-weighted Class Activation Mapping (Grad-CAM), SHapley Additive exPlanations (SHAP), Local Interpretable Model-agnostic Explanations (LIME), and attention visualization mechanisms. We provide detailed mathematical formulations, implementation strategies, and comparative analyses of these approaches. Our investigation reveals that while Grad-CAM excels at speed and computational efficiency, SHAP provides theoretically sound explanations grounded in cooperative game theory. LIME offers model-agnostic interpretability, and attention mechanisms provide inherent explainability through learned attention weights. Furthermore, we discuss applications in high-stakes domains including medical imaging, autonomous driving, and facial recognition, where explainability is not merely beneficial but legally and ethically mandated. This survey provides practitioners and researchers with comprehensive guidance for selecting and implementing appropriate explainability methods for their specific applications.},
      keywords = {Explainability, Interpretability, Grad-CAM, SHAP, LIME, Attention Mechanisms, Deep Learning, Computer Vision, Neural Network Interpretability, Feature Importance},
      month = {September},
  }