A simple REST API server built using the gotham web framework, tract for neural network inference, and Rust.
This project was created as a demo for an accompanying talk: Offline ML with Rust. The slide deck is included in this repo in the /docs folder.
- Before running the code, you will need to download the model:
wget https://storage.googleapis.com/mobilenet_v2/checkpoints/mobilenet_v2_1.4_224.tgz
tar zxf mobilenet_v2_1.4_224.tgz
Most of the downloaded material is junk, we only care about the mobilenet_v2_1.4_224_frozen.pb file.
- Build and start the server:
cargo run
- The API server listens for POST requests to the root endpoint with a body containing a serializable image (I've only tested with JPEGs). You can use cURL to post images to the server:
curl -i -X POST -F "image=@<IMAGE_PATH>" http://127.0.0.1:7878/
The returned value is a tuple: (score, class). Score is a value between 0 and 1 signifying the model's confidence that the given class is in the image. Class is the line number associated with the predicted class in the classes text file.