🌌 ConvNet-Vis

ConvNet-Vis helps to visualize the Deep Convolutional Neural Networks with following methods.

• Activation of image from each layer
• Deconvolution
• Deep-Dream

Requirements:

• Tensorflow
• Keras
• numpy
• scipy
• h5py
• wget
• six
• scikit-image

Run:

Very first time it will download the weights of the model you pick, so it requires an internet connection. Also you can pass custom model.

import vis

# For pretrained model visualization
convnet_vis = vis.ConvNet_Vis(image_path="cat.jpg")

# For custom model visualization
# model=<path-to-model> OR Keras Model obect
convnet_vis = vis.ConvNet_Vis(image_path="cat.jpg", model="final_model.hdf5")

NOTE: Results will be stored in vis directory.

Google-Colab support added.

For Tensorboard:

TensorBoard gives you flexibility to visualize all the test image on same model with brightness and contrast adustment.

tensorboard --logdir=<LOG-PATH (layerwise)>

Todo:

• Deep-dream support for all the model. (Current Support only for InceptionV3)

License

MIT

Author

Acknowledgement

💚 tfcnn_vis