Skip to content

hrishikeshkh/XCNN

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

27 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Deconvolve: Towards Textually Explainable and Human Cognizable Convolutional Networks

The repository helps reproduce the results of our paper, under consideration for publication in the IJEECS.

Description:
This project focuses on analyzing the CIFAR-10 dataset using a pre-trained CNN model. It includes functionality for generating Grad-CAM heatmaps to visualize feature activations and explain model predictions.


Table of Contents

  1. Introduction
  2. Features
  3. Installation
  4. Usage
  5. Contact

Introduction

Convolutional Neural Networks (CNNs) have demonstrated exceptional accuracy in image classification and are widely used across various industries. However, understanding their decision-making process remains a challenge due to their black-box nature. To enhance model interpretability, we introduce DeConvolve—a post-hoc explainability methodology that leverages multimodal Large Language Models (LLMs) to extract human-understandable insights from CNN feature maps.


Features

  • CIFAR-10 Dataset Filtering: Filter the dataset to include specific classes (e.g., airplane, automobile, bird).
  • Model Prediction: Use a pre-trained CNN model to classify images.
  • Grad-CAM Heatmaps: Generate heatmaps to visualize feature activations.
  • Neuron Influence Analysis: Analyze the influence of neurons on predictions.
  • Explainability: Generate human-readable explanations for model predictions.

Installation

Clone the repository and Install dependencies:

pip install -r requirements.txt

Usage

Please refer to the file structure below, once the required packages are installed, the notebooks help us walkthrough the proposed method.

The order to run the files upon setup:

To run again:

File Structure

[Project Root Directory]
│
├── assets/                   # Directory for heatmaps and images
├────[Heuristic.ipynb]        # Heuristic code
├────[pneumonia.ipynb]        # Preprocessing
├── [copy_of_cifar.py]        # Main script for CIFAR-10 analysis
├── [requirements.txt]        # Dependencies
├── [README.md]               # Project documentation
└── [Other files...]

Contact

For any questions or feedback, please contact: hrishikeshh[at]gmail[dot]com

Copyright: Hrishikesh K Haritas, Darshan Bankapure, Divyansh Mishra under the Apache 2.0 license.


About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 3

  •  
  •  
  •