Deshram / dockerize-image-matching-model

Dockerizing a deep learning model

Dockerize Deep learning model

The objectibe of this repository is to give overview of deplyoing a pretrained deep learning computer vison model by dockerizing it, which is based on kornia framework which matches the given two image.

Kornia is a computer vision framework built on top of pytorch

The pretrained model which has been used is kornia implementation of LoFTR: Detector-Free Local Feature Matching with Transformers. LoFTR can extract high-quality semi-dense matches even in indistinctive regions with low-textures, motion blur, or repetitive patterns.

To-Do

1. Install Docker as per your operating system. Ignore this part if you already have it.
2. Clone this repository by running git clone https://github.com/Deshram/dockerize-image-matching-model
3. Make a models folder and download weights from below links
4. Here I've used outdoor weights, but you can also try indoor and indoor new weights which are trained on different dataset by downloading them in models folder
5. Build docker image by running docker build -t IMAGE_NAME . e.g. docker build -t image_mathcing .
6. Run docker container to deploy on host docker run --gpus all -p HOST_PORT:CONTAINER_PORT IMAGE_NAME e.g. docker run --gpus all -p 5000:5000 image_matching PS: Use --gpus all if host machine contains GPU

demo_1.mp4