CNN可视化 Feature Map 的四种方法

参考论文：《How convolutional neural network see the world - A survey of convolutional neural network visualization methods》

• Activation Maximization：
• Network Inversion：
• Deconvolutional Neural Networks (DeconvNet)：
• Network Dissection：

后面有相关的参考论文。

如果是想要可视化CNN的网络结构，推荐使用这个软件：Netron

Activation Maximization：

用BP的算法生成一个输入图像，该图像可以最大化任意层中特定神经元的激活。
三种方式：

• 不带正则：
  $$x^{\ast }=\underset{x}{\mathrm{argmax}}{a}_{i,l}(\theta ,x)$$
• 带正则：
  $$x^{\ast }=\underset{x}{\mathrm{argmax}}(a_{i,l}(\theta ,x)-\lambda (x))$$
• 与GAN结合（Deep Generative Network Activation Maximization, DGN-AM）：
  $$x^{\ast }=\underset{x}{\mathrm{argmax}}(a_{i,l}(\theta ,G(x))-\lambda (x))$$

Network Inversion：

在输入的图像中找到一个特定的区域，这个区域能够使卷积层中特定的神经元激活。

• 基于Regularizer：
  $$x^{\ast }=\underset{x}{\mathrm{argmax}}(C\cdot \mathcal{L}(A(x),A(x_0))-\lambda (x))$$
• 基于UpconvNet：
  $$W^{\ast }=\underset{w}{\mathrm{argmax}}\sum _i||x_i-D(A(x_i),W)|{|}^2$$
  

Deconvolutional Neural Networks (DeconvNet)：

对于输入图像，通过任意层中所有神经元的特征图重建新图像，以突出全面的CNN层级特征。
可以将多个特征图$a_{i,l}$转化为一维特征向量$A_l$，而原来的卷积操作可视为向量与向量之间的映射$f_l$：
$$A_{l+1}=A_l\ast f_l$$
所以Deconv的操作是如下的方式：
$$R_l=R_{l+1}\ast f_l^T$$

还有其他两种操作：$ReversedRectificationLayer$ 与 $ReversedMax-pooling/UnpoolingLayer$
参考：https://blog.csdn.net/qq_16234613/article/details/79387345

Network Dissection：

评估单个或多个卷积神经元与特定语义概念之间的相关性。
项目代码：https://github.com/CSAILVision/NetDissect
官方主页：http://netdissect.csail.mit.edu/

---

参考文献：

• Activation Maximization：
  • D. Erhan, Y. Bengio, A. Courville and P. Vincent, Visualizing higher-layer features of a deep network, Technical report, University of Montreal, 1341 (2009), p3.
  • G. E. Hinton, S. Osindero and Y.-W. Teh, A fast learning algorithm for deep belief nets, Neural computation, 18 (2006), 1527-1554.
  • P. Vincent, H. Larochelle, I. Lajoie, Y. Bengio and P.-A. Manzagol, Stacked denoising autoencoders: Learning useful representations in a deep network with a local denoising criterion, Journal of Machine Learning Research, 11 (2010), 3371-3408.
  • K. Simonyan, A. Vedaldi and A. Zisserman, Deep inside convolutional networks: Visualising image classi cation models and saliency maps, arXiv preprint, arXiv:1312.6034.
  • J. Yosinski, J. Clune, A. Nguyen, T. Fuchs and H. Lipson, Understanding neural networks through deep visualization, arXiv preprint, arXiv:1506.06579.
  • A. Nguyen, J. Yosinski and J. Clune, Multifaceted feature visualization: Uncovering the different types of features learned by each neuron in deep neural networks, arXiv preprint, arXiv:1602.03616.
  • A. Nguyen, A. Dosovitskiy, J. Yosinski, T. Brox and J. Clune, Synthesizing the preferred inputs for neurons in neural networks via deep generator networks, in Proceedings of the Advances in Neural Information Processing Systems, 2016, 3387-3395.
• Network Inversion：
  • A. Mahendran and A. Vedaldi, Understanding deep image representations by inverting them, in Proceedings of the IEEE conference on Computer Vision and Pattern Recognition, 2015, 5188-5196.
  • A. Mahendran and A. Vedaldi, Visualizing deep convolutional neural networks using natural pre-images, International Journal of Computer Vision, 120 (2016), 233-255.
• Deconvolutional Neural Networks (DeconvNet)：
  • M. D. Zeiler and R. Fergus, Visualizing and understanding convolutional networks, in Proceedings of the European Conference on Computer Vision, 2014, 818-833.
• Network Dissection：
  • D. Bau, B. Zhou, A. Khosla, A. Oliva and A. Torralba, Network dissection: Quantifying interpretability of deep visual representations, in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017, 3319-3327.
  • R. Fong and A. Vedaldi, Net2vec: Quantifying and explaining how concepts are encoded by lters in deep neural networks, arXiv preprint, arXiv:1801.03454.