Please use Turbowarp with hqpen and double click green flag: https://turbowarp.org/1322285624?hqpen Please hover your mouse over the images from the conv layers and above (not the input layer image). Model Used: https://adamharley.com/nn_vis/harley_vis_isvc15.pdf Input Layer: Contains 1024 nodes on the bottom layer. While the network classifies 28 × 28 28×28 pixel images, the 1024 nodes indicate the input is mapped to a 32 × 32 32×32 grid. Convolution Layer 1: Utilises 6 filters of size 5 × 5 5×5 operating with a stride of 1. Downsampling Layer 1: Applies 2 × 2 2×2 max pooling with a stride of 2. Convolution Layer 2: Utilises 16 filters of size 5 × 5 5×5 operating with a stride of 1. Downsampling Layer 2: Again applies 2 × 2 2×2 max pooling with a stride of 2. Fully-Connected Layers: Three dense layers in succession containing 120 nodes, 100 nodes, and a final output layer of 10 nodes (representing the prediction for digits 0-9).
Paper used: https://adamharley.com/nn_vis/harley_vis_isvc15.pdf The model used as well as the original project from. https://adamharley.com/nn_vis/cnn/2d.html It is really incredible, he also has other visualisations in 3D as well. I highly recommend for people to check them out, they are really fun to use and super intuitive. p.s. I wont be posting for a while as I have A level mocks, so I wont be online for quite a while again :/
