API for grouping images on similarity.
Optional you can sample your image set based on clusters. Sampling your images makes sense if you have a hugh amount of them and wanna label only a subset by using an image labeling tool.
- docker
- docker-compose
To check if docker-ce is installed:
docker --version
To check if docker-compose is installed:
docker-compose --version
Ubuntu
To install Docker and Docker Compose on Ubuntu, please follow the link.
Windows 10
To install Docker on Windows, please follow the link.
In order to build the project run the following command from the project's root directory:
sudo docker-compose up --build
To see all the available endpoints, open your favorite browser and navigate to:
http://<machine_IP>:5002/docs
You can change the port in the docker compose file.
/download_web_images (POST)
This endpoint takes a search query and downloads corresponding images from google, bing and yahoo.
Images will be saved to the mounted input directory, which can be changed in the docker compose file:
├──exchange/
├── input/
├── *.jpg
├── *.jpeg
├── *.png
The endpoint returns an overview of all downloaded images. For example when searching for 'germany':
/get_amount_of_clusters (POST)
This endpoint takes the image set from the mounted input directory and returns the result of the elbow method for choosing the best amount of clusters:
Please place files in the mounted input directory.
├──exchange/
├── input/
├── *.jpg
├── *.jpeg
├── *.png
/cluster_images (POST)
This endpoint takes the image set from the mounted input directory and groups them based on similarity.
In addition you have the opportunity to get a sample of your image set by choosing x percentage of each image cluster.
Please place files in the input directory. The output directory will be generated automaticly:
├──exchange/
├── input/
│ ├── *.jpg
│ ├── *.jpeg
│ ├── *.png
│
├──output/
├── cluster_0/
│ ├── *.jpg
│ ├── *.jpeg
│ ├── *.png
├── cluster_1/
│ ├── *.jpg
│ ├── *.jpeg
│ ├── *.png
├── sample/
├── *.jpg
├── *.jpeg
├── *.png
The endpoint returns an overview of the image clusters: