卷积神经网络可视化与可解释性

• 对卷积网络可视化与可解释性相关资料的一些整理，不断更新中~
  最新更新参照知乎

---

主要作用

• 进一步理解CNN
• 帮助我们分析训练好的网络
• 学会如何生成热力图，如下图:

---

博客

• Distill 非常推荐的一个网站
• Fun With ConvNets (dataviz!)
• Global Average Pooling Layers for Object Localization
• Yes, Deep Networks are great, but are they Trustworthy?
• 凭什么相信你，我的CNN模型？

---

论文

• Visualizing and Understanding Convolutional Networks
• Striving for Simplicity: The All Convolutional Net
• Learning Deep Features for Discriminative Localization
• Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localization
  

---

代码

• Keras作者《Deep learning with python》一书中关于可视化方法的ipynb
  • 可视化卷积层的输出
  • 可视化卷积核
  • 生成heatmap–grad-CAM方法
  • 《Deep learning with python》下载地址
• 摘自https://github.com/mrgloom/CNN-heatmap

1. Visualizing and Understanding Convolutional Networks
https://arxiv.org/pdf/1311.2901v3.pdf
https://neukom.dartmouth.edu/docs/bbat-wacv2016.pdf
http://cs231n.github.io/understanding-cnn/

2. Net surgery trick
http://cs231n.github.io/convolutional-networks/#convert
https://github.com/BVLC/caffe/blob/master/examples/net_surgery.ipynb

https://leonardoaraujosantos.gitbooks.io/artificial-inteligence/content/image_segmentation.html
https://arxiv.org/pdf/1502.02766v3.pdf
https://arxiv.org/pdf/1411.4038v2.pdf
http://www.cv-foundation.org/openaccess/content_cvpr_2015/papers/Long_Fully_Convolutional_Networks_2015_CVPR_paper.pdf
https://arxiv.org/pdf/1605.06211v1.pdf

`convert fully connected layers to their
    equivalent convolutional layers, since the weights are the same and only
    the shapes are different.`

3. Global average pooling layer
Network In Network
https://arxiv.org/pdf/1312.4400.pdf

4. Learning Deep Features for Discriminative Localization
http://cnnlocalization.csail.mit.edu/
https://github.com/jacobgil/keras-cam

5. Grad-CAM: Gradient-weighted Class Activation Mapping
https://github.com/ramprs/grad-cam
http://gradcam.cloudcv.org/
https://arxiv.org/pdf/1610.02391v2.pdf

6. Is object localization for free? – Weakly Supervised Object Recognition with Convolutional Neural Networks
http://www.di.ens.fr/willow/research/weakcnn/
http://www.di.ens.fr/willow/research/cnn/

Additional:

1. Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps
https://arxiv.org/pdf/1312.6034v2.pdf

2. Top-down NeuralAttention by Excitation Backprop (c-MWP)
https://arxiv.org/pdf/1608.00507v1.pdf
http://cs-people.bu.edu/jmzhang/excitationbp.html

3. Paying More Attention to Attention: Improving the Performance of Convolutional Neural Networks via Attention Transfer
https://arxiv.org/pdf/1612.03928v1.pdf
https://github.com/szagoruyko/attention-transfer

4. Visualizing Deep Neural Network Decisions: Prediction Difference Analysis
https://openreview.net/pdf?id=BJ5UeU9xx
https://github.com/lmzintgraf/DeepVis-PredDiff
A New Method to Visualize Deep Neural Networks
https://icmlviz.github.io/assets/papers/23.pdf

5. Self-Taught Object Localization with Deep Network
https://arxiv.org/pdf/1409.3964.pdf

6. Shallow and Deep Convolutional Networks for Saliency Prediction
https://github.com/imatge-upc/saliency-2016-cvpr

Keras codebase:

https://github.com/raghakot/keras-vis

Oclusion based technique:
https://github.com/waleedka/cnn-visualization/blob/master/cnn_visualization.ipynb

Net surgery trick:
https://github.com/heuritech/convnets-keras

GAP-CAM
https://github.com/alexisbcook/ResNetCAM-keras/
https://github.com/jacobgil/keras-cam
https://github.com/tdeboissiere/VGG16CAM-keras

Grad-CAM
https://github.com/jacobgil/keras-grad-cam

https://raghakot.github.io/keras-vis/visualizations/attention/

https://github.com/mlhy/ResNet-50-for-Cats.Vs.Dogs

Tensorflow codebase:

GAP
https://github.com/sjchoi86/tensorflow-101/blob/master/notebooks/gap_mnist.ipynb

GAP-CAM
https://github.com/jazzsaxmafia/Weakly_detector

Grad-CAM
https://github.com/Ankush96/grad-cam.tensorflow

PyTorch

Grad-CAM
https://github.com/kazuto1011/grad-cam-pytorch

GAP-CAM
https://github.com/metalbubble/CAM/blob/master/pytorch_CAM.py

SPN
https://github.com/yeezhu/SPN.pytorch

https://github.com/utkuozbulak/pytorch-cnn-visualizations

https://github.com/jacobgil/pytorch-explain-black-box

Regression Activation Map

https://github.com/cauchyturing/kaggle_diabetic_RAM

Other

https://github.com/metalbubble/cnnvisualizer

https://github.com/InFoCusp/tf_cnnvis

https://github.com/ppwwyyxx/tensorpack/tree/master/examples/Saliency

https://github.com/InFoCusp/tf_cnnvis

https://github.com/keplr-io/quiver

https://jacobgil.github.io/deeplearning/vehicle-steering-angle-visualizations
https://jacobgil.github.io/deeplearning/class-activation-maps

https://jacobgil.github.io/deeplearning/filter-visualizations
https://jacobgil.github.io/computervision/saliency-from-backproj

https://github.com/CSAILVision/NetDissect

https://medium.com/merantix/picasso-a-free-open-source-visualizer-for-cnns-d8ed3a35cfc5

https://github.com/CSAILVision/NetDissect
http://netdissect.csail.mit.edu/

https://github.com/imatge-upc/saliency-salgan-2017

http://imatge-upc.github.io/saliency-2016-cvpr/
https://github.com/imatge-upc/saliency-2016-cvpr

Visualization:

https://github.com/shaohua0116/Activation-Visualization-Histogram

To look at:

http://blog.qure.ai/notes/visualizing_deep_learning