ImageSearchProject

Instruction

Install requirements

1. Creating coda environtment:

conda create --name py38 python==3.8.16
conda avtive py38

3. Install Pytorch-cuda==11.7, following official instruction:

conda install pytorch torchvision torchaudio pytorch-cuda=11.7 -c pytorch -c nvidia

3. Install the necessary dependencies by running:

pip install -r requirements.txt

Preparing datasets

1. Downloading Oxford5k datasets in here.
2. Downloading Paris6k datasets in here
3. Put the dataset following this structure:

ImageSearchProject
│
├── static/
│ ├── oxford
| | ├── feature
| | | ├── LBP.index.bin
| | | ├── Resnet50.index.bin
| | | ├── VGG16.index.bin
| | | ├── RGBHistogram.index.bin
| | | ├── ViT.index.vin
| | ├── evaluation
| | | ├── crop
| | | ├── original
| | ├── groundtruth
| | ├── image
| | | ├── all_souls_000000.jpg
| | | ├── all_souls_000001.jpg
| | | └── ...
| ├── paris
| | ├── feature
| | | ├── LBP.index.bin
| | | ├── Resnet50.index.bin
| | | ├── VGG16.index.bin
| | | ├── RGBHistogram.index.bin
| | | ├── ViT.index.vin
| | ├── evaluation
| | | ├── crop
| | | ├── original
| | ├── groundtruth
| | ├── image
| | | ├── paris_defense_000000.jpg
| | | ├── paris_defense_000042.jpg
| | | └── ...
└── ...

Running code

1.Indexing (feature extraction)

python indexing.py --data_path static/oxford --feature_extractor resnet50

The resnet50.index.bin will be at static/oxford/feature 2. Ranking

python ranking.py --data_path static/oxford --feature_extractor resnet50 --k 10

3. Evaluation

python evaluation --data_path static/oxford --feature_extractor resnet50

Running demo

• Using flask to build web
• Running command line below to start:

flask run

Reference

@article{johnson2019billion,
title={Billion-scale similarity search with {GPUs}},
author={Johnson, Jeff and Douze, Matthijs and J{\'e}gou, Herv{\'e}},
journal={IEEE Transactions on Big Data},
volume={7},
number={3},
pages={535--547},
year={2019},
publisher={IEEE}
}