Object localization

This repository trains model to perform object localization task on Pets Dataset and visualizes the CNN filters.

Interpretation methods available

1. Guided Backpropagation (GBP)
2. Integrated Gradients (IG)
3. Smooth-Guided Backpropagation
4. Smooth-Integrated Gradients
5. Grad-CAM
6. Activation Maximization
7. Local Interpretable Model-agnostic Explanations (LIME)
8. DeepExplainer - SHapley Additive exPlanations (SHAP) + DeepLIFT
9. GradientExplainer - SHAP + IG + SmoothGrad

References

[1] https://keras.io/examples/vision/visualizing_what_convnets_learn/

[2] https://jacobgil.github.io/deeplearning/filter-visualizations

[3] https://www.kaggle.com/vincentman0403/visualize-output-of-cnn-filter-by-gradient-ascent

[4] https://medium.com/@jon.froiland/convolutional-neural-networks-part-8-3ac54c9478cc

[5] https://github.com/sicara/tf-explain#available-methods

[6] Cristian Vasta, Deep Learning Models with Tensorflow 2.0, Available at: https://morioh.com/p/64064daff26c, Accessed on: 03. 05. 2021.

[7] Hoa Nguyen, CNN Visualization Keras TF2, GitHub, Available at: https://github.com/nguyenhoa93/cnn-visualization-keras-tf2, Accessed on: 03. 05. 2021.

[8] Available at: https://pair-code.github.io/saliency/#guided-ig