VGG16-Image-Reverse-Engine

An Image Reverse Engine based on VGG16

Introduction

This is a project for the course "Application of artificial intelligence and deep learning" in the National Kaohsiung University of Science and Technology. The goal of this project is to build an image reverse engine based on VGG16.

Requirements

pip install -r requirements.txt

Usage

main.ipynb

The main.ipynb is the main file of this project. You can run it on Jupyter Notebook or Google Colab. It contains the following parts:

1. Load data
2. Load VGG16 model
3. Indexing images
4. Save features and indexes
5. Load features and indexes
6. Search images
7. Show results

The dataset we used is from Stanford University CS231n. You can download it and put it in the same directory as the main.ipynb.

query.ipynb

The query.ipynb is a file for you to query images. You can run it on Jupyter Notebook or Google Colab. It contains the following parts:

1. Search images without any modification
2. Search images with Gaussian Noise
3. Search images with clustering (K-Means) and Gaussian Noise

This file also explains the process of searching images in detail.

indexing.py

The indexing.py is a file for you to index images. It is separated from the index part in main.ipynb.

model.py

The model.py is a file for you to load VGG16 model. You can use it to load VGG16 model in other files.

from model import VGGNet
model = VGGNet()

and then you can use the model to extract features.

features = model.extract_feature(image)

web_app.py

The web_app.py is a file for you to run a web application built by Gradio. You can run it on your local machine.

python web_app.py

Reference

Tiny Imagenet 200 from Stanford University CS231n

how does the dot product determine similarity?

Image Similarity Comparison using VGG16 Deep Learning Model

基于VGG-16的海量图像检索系统（以图搜图升级版）

点积相似度、余弦相似度、欧几里得相似度