Profiling-Vision-Neural-Networks-with-CLIP-DISSECT

This is the Codebase for "Profiling Vision Neural Networks with CLIP-DISSECT" report.

Installation

1. Install Python (3.10)
2. Install remaining requirements using pip install -r requirements.txt

Quick Start

run python network_profiling.py

neuron descriptions from CLIP-DISSECT of ResNet-18/50, and ViT B/16 are provided in this repo.

How to modify

1. run with your own subject model: following CLIP-DISSECT to dissect your own vision model, get neuron explanations. Then, run python network_profiling.py --neuron_explanation_path [YOUR OWN NEURON EXPLANATIONS PATH] --NN_type [YOUR MODEL NAME] --interpretable_threshold [YOUR UNINTERPRETABLE THRESHOLD] --categories [YOUR CATEGORIES]

2. replace y axis as neuron counts rather than percentage: python network_profiling.py --unnormalize