Currently the repo does not have a license file, it should be added before initial release.

Need a simple README before releasing the first version

Since the package does not have a description publishing failed here: https://github.com/kayagokalp/handtrack-rs/actions/runs/3777803543/jobs/6421821307

Currently the input image is used fed to the model directly. It could make sense to downsize the image first especially for the performance critical usage of this library (detection from real-time video).

Currently the input image is used fed to the model directly. It could make sense to re-color (to grayscale?) the image first especially for the performance critical usage of this library (detection from real-time video). I am not sure if this would improve performance of the detection speed wise and it should be tested first. This could also effect the detection performance.

I forgot to update the readme with the change introduced in #19

Create the MVP API design and got to the point we can run simple detection queries through the API.

Since the permissions are not correctly set, verify tag script cannot be executed which caused failure in this run: https://github.com/kayagokalp/handtrack-rs/actions/runs/3777768976

Add CI checks for:

1. Running cargo test.
2. Checking formatting violations.
3. Checking linter warnings.
4. Publishing step

As of now, I am only testing things with single images. Eventually handtrack-rs should be able to detect hands from a video stream. This probably requires #14 to be tackled first. It would be great to have some tests for video stream detection and provide some benchmark in the README.

I wrote in a hurry :) and it can use some paraphrasing and little touch-ups.

ChatGPT provides the following output for this request:

handtrack-rs

Welcome to handtrack-rs, a Rust library for detecting hands in images! This library is inspired by handtracking and aims to provide a simple, easy-to-use API for detecting hands in real-time video.

Example

Here's an example of how to use the detect function:

let image = Image::from_file(project_dir).unwrap();

let score_threshold = 0.7f32;
let max_hands = 1;
let detection_opts = DetectionOptions::new(max_hands, score_threshold);

let detection = detect(image, detection_opts).unwrap();

let detection_box = &detection[0];

The detect function takes in an Image and DetectionOptions, allowing you to specify the maximum number of hands to detect and a score threshold for classifying an object as a hand.

Contribution

This library is still a work in progress and contributions are welcome! To contribute, please make sure to run cargo clippy, cargo fmt, and cargo test checks and ensure that Cargo.toml dependencies are in alphabetical order.

I think it looks good but will leave this open for now and update the README once I have more features.

Since the graph to be embedded is bigger than 10 Mb (which is the crates.io's limit), uploading a package to crates.io is not possible atm.

Happened at: https://github.com/kayagokalp/handtrack-rs/actions/runs/3777837940

Currently detect() directly uses frozen inference graph and does not allow users to specify the model themselves. This is not ideal as it means reading the model for each detection. This can be improved greatly by simply taking the model loading/creating part out of the detect() function and accept it as a parameter.