Model Viewer

This application will show the output of each layer of a Convolution Neural Network (CNN). You can select the layer you want and the output will be shown. (If the viewing space is not enough, the images will split into pages)

Current there is only 2 models support this preview, only Densenet161 and ResNet50.
For DenseNet161, I used the code available here. You can download the model from there and test the program.

You can add custom models also. What you will need to do is adding the model file inside the model package, add the model details to the lib/common.py and add how to call your models in libs/feature_extractor.py file. Also in ui/main_window.py file, edit the model name given to the feature extractor.
Make sure you have public model variable with with Keras model and preprocess() function in you custom model class. (Look how I have created the densenet and resnet models for more information)

Changing Settings

You can change the settings of the program by editing lib/common.py file.

image_path variable	Location of the images. Images should be in .jpg format.
MODEL_TYPE class	Each model in the application has to add there. If you add a new model, create a variable here.
model_locs variable	Location of each trained Keras Model provided in MODEL_TYPE class has to list here.
Sizes class	You can change the image sizes of the visualization from here.

• querry_x and querry_y: Size of the input image shown.
• image_x and image_y: Size of the layer outputs.
• n_rows: Number of rows displayed in a single page.
• n_cols: Number of columns displayed in a single page.

Run the program

You can the the program by

python main.py

Output shown by some layers when it was running with DenseNet161 model is shown below.