Example use of image classification using Vision framework

ImageClassificationService.swift

import Foundation
import FuntastyKit
import PromiseKit
import Photos
import Vision
import QuartzCore

protocol ImageClassificationService: Service {
    func classify(asset: PHAsset) -> Promise<String>
}

final class VisionImageClassificationService: ImageClassificationService {

    private let model: VNCoreMLModel = try! VNCoreMLModel(for: Resnet50().model)

    private let modelInputResolution: CGFloat = 224

    private var modelInputSize: CGSize {
        return CGSize(width: modelInputResolution, height: modelInputResolution)
    }

    func classify(asset: PHAsset) -> Promise<String> {
        return Promise { fullfill, reject in
            PHImageManager.default().requestImage(for: asset, targetSize: modelInputSize, contentMode: .aspectFill, options: options) { image, info in
                let request = VNCoreMLRequest(model: self.model) { request, error in
                    if let error = error {
                        reject(error)
                    } else {
                        let observations = request.results as? [VNClassificationObservation] ?? []
                        fullfill(observations.first?.identifier ?? "Picture")
                    }
                }
                let handler = VNImageRequestHandler(cgImage: image!.cgImage!)
                try! handler.perform([request])
            }
        }
    }

    private var options: PHImageRequestOptions {
        let options = PHImageRequestOptions()
        options.isSynchronous = true
        options.deliveryMode = .fastFormat
        options.resizeMode = .exact
        options.normalizedCropRect = CGRect(origin: .zero, size: modelInputSize)
        return options
    }
}