Gradient-Ascent-Feature-Visualization

Investigating and cross-analyzing class-specific feature visualizations across pretrained Convolutional Neural Networks (VGG-16, VGG-19, and ResNet-152) using gradient ascent. This project explores how optimization techniques, specifically $L_2$ regularization and Gaussian blur smoothing impact visual interpretability and cross-model transferability.

For the complete problem definition, methodology, and implementation details, refer to the Jupyter Notebook.

Environment Setup

This project provides Conda environment configurations for CPU and GPU setups across Windows and Unix-based systems. The following assumes Conda is installed on your system.

1. Clone the Repository

git clone https://github.com/O-Brob/Gradient-Ascent-Feature-Visualization.git
cd Gradient-Ascent-Feature-Visualization

2. Create the Conda Environment

Select the configuration file from the conda-environment-files/ directory that matches your operating system and hardware setup. e.g. if using Linux with a GPU: conda env create -f conda-environment-files/conda-environment-gpu-unix.yml where gpu is interchangeable with cpu and unix with win.

3. Activate the Environment

conda activate dml

4. Launch Jupyter Notebook

jupyter notebook