Recommend a Convolutional Neural Network (Convolutional Neural Network) visualization website https://poloclub.github.io/cnn-explainer/ .
Researchers from Georgia Institute of Technology and Oregon State University have jointly developed a convolutional neural network interactive visualization tool-CNN Explainer. This interpreter shows a 10-layer neural network, including convolutional layers, activation functions, pooling layers and many other concepts that CNN beginners can't avoid anyway.
CNN Explainer uses TensorFlow.js to load the pre-trained model for visualization, and the interaction uses Svelte as the framework and D3.js for visualization. The final product has no barriers to use even for novices who do not understand.
If you hover the mouse over the top activation map of the first convolutional layer, you can see that 3 convolution kernels are applied here to get this activation map. Click on this activation map and you can see that each convolution kernel has performed convolution operations.
The following is the convolution kernel used to generate the intermediate result at the top of the above activation map.
You can also demonstrate the role of hyperparameters, and there are a series of functions waiting for everyone to explore.
reference
Project address: https://github.com/poloclub/cnn-explainer
web address: https://poloclub.github.io/cnn-explainer/
arXiv address: https://arxiv.org/abs/2004.15004
https:/ /baijiahao.baidu.com/s?id=1665923868483719652&wfr=spider&for=pc