Human action classification for video, offline and natively on iOS via Core ML

Uses the kinetics-i3d model to classify videos into one of 400 different action classes defined in Kinetics 400

⚠️ This project requires Xcode 12.

Reference: See accompanying blog post

To install via Swift Package Manager, add VisualActionKit to your Package.swift file. Alternatively, add it from Xcode directly.

let package = Package(
    ...
    dependencies: [
        .package(url: "https://github.com/lukereichold/VisualActionKit.git", from: "0.1.0")
    ],
    ...
)

Then import VisualActionKit wherever you’d like to use it:

import VisualActionKit

let url = Bundle.module.url(forResource: "writing", withExtension: "mp4")
let asset = AVAsset(url: url)

try Classifier.shared.classify(asset) { predictions in
    print(predictions)
}

Contributions welcome. Please check out the issues.