Keras implementation to visualize outputs and weights of fully connected layer.
This repository contains keras (tensorflow.keras) implementation to visualize outputs and weights of fully connected layer of common CNN (VGG8) and ArcFace [1] using Fashion MNIST dataset [1].
Fashion MNIST dataset includes 10 classes in 4 categories:
- Tops: T-shirt/top, Pullover, Dress, Coat, Shirt
- Bottoms: Trouser
- Shoes: Sandal, Sneaker, Ankle boot
- Bags: Bag
Figure 1 show the output vectors of the fully connected layer of VGG8 trained on Fashion MNIST dataset. Figure 2 show the weight vectors of the fully connected layer of VGG8 trained on Fashion MNIST dataset.
Figure 1. Output of fully connected layer in (a) train and (b) test data
Figure 2. Weight of fully connected layer
Figure 3 show the output vectors of the fully connected layer of VGG8 trained on Fashion MNIST dataset. Figure 4 show the weight vectors of the fully connected layer of VGG8 trained on Fashion MNIST dataset.
Figure 3. Output of fully connected layer in (a) train and (b) test data
Figure 4. Weight of fully connected layer
[1] Deng, J., Guo, J., Xue, N., & Zafeiriou, S. (2019). Arcface: Additive angular margin loss for deep face recognition. CVPR.
[2] Xiao, H., Rasul, K., & Vollgraf, R. (2017). Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms. arXiv preprint arXiv:1708.07747.
[3] https://github.com/4uiiurz1/keras-arcface