zjykzj / crow-pytorch

[ECCV 2016] Cross-dimensional weighting for aggregated deep convolutional features.

Language: 🇺🇸 🇨🇳

«crow-pytorch» uses Pytorch to reproduce the CroW implementation.

Table of Contents

• Table of Contents
• Background
• Installation
• Usage
• Maintainers
• Thanks
• Contributing
• License

Background

CroW provides a general convolution feature extraction framework, and proposes parameterless spatial weighting and channel weighting algorithms. In addition, a very detailed implementation is provided - YahooArchive/crow.

The official implementation is based on caffe2, but the most popular deep reasoning framework at present is pytorch. In order to better understand the implementation of CroW, I try to replace the implementation of caffe in the warehouse with pytorch.

Installation

pip install -r requirements.txt

Usage

1. Get data

bash oxford/get_oxford.sh
bash paris/get_paris.sh

2. Extract features

python extract_features.py --images oxford/data/* --out oxford/layer4 --layer layer4
python extract_features.py --images paris/data/* --out paris/layer4 --layer layer4
python extract_queries.py --dataset oxford --images data --groundtruth groundtruth --layer layer4

3. Compile eval tool

g++ -O compute_ap.cpp -o compute_ap

4. Evaluate

python evaluate.py --queries oxford/layer4_queries --groundtruth oxford/groundtruth --index_features oxford/layer4 --wt crow --dw 3 --whiten_features paris/layer4 --d 512 --qe 3

Maintainers

• Clayton Mellina - Initial work - pumpikano
• zhujian - Enhance work - zjykzj

Thanks

• YahooArchive/crow

Contributing

Anyone's participation is welcome! Open an issue or submit PRs.

Small note:

• Git submission specifications should be complied with Conventional Commits
• If versioned, please conform to the Semantic Versioning 2.0.0 specification
• If editing the README, please conform to the standard-readme specification.

License

Apache License 2.0 © 2022 zjykzj