Deep Convolutional Neural Networks (DCNNs) achieve impressive results on image Classification, object detection and object tracking tasks. However, researchers still do not clearly understand what is the network learned during training phase. Recent works shows the state-of-the-art result on visualization of DCNNs. However, DCNNs are still a ”black-box” for most of beginners and other field researchers. In order to improve the interpretability and make the network easy to understand, I propose a manner to easily visualize the output of each layer. As part of the work, I also plan to conduct ablation study on how neural net work works and what it may learned during training.
Before running the code, download the Tiny ImageNet data set in the official website
1. Following the guide on tiny ImageNet and put the data under TinyImageNet folder as shown in the example folder
2. To train the model use "python3 train.py" the output model will be under models folder
3. To test the model use "python3 test.py"