Implement Image Classification Using a Trained Core ML Model and Vision Framework

The Product Classification Table View Controller is supposed to display products of the classified product category. You have implemented similar code before so this should look familiar to you.

Add the needed import statements below the import UIKit statement above the class declaration:

import SAPFiori
import SAPOData
import SAPCommon

Add the following properties right above the viewDidLoad(_:) method:

private let logger = Logger.shared(named: "ProductClassificationTableViewController")

// The available destinations from Mobile Services are hold in the FileConfigurationProvider. Retrieve it to find the correct data service
let destinations = FileConfigurationProvider("AppParameters").provideConfiguration().configuration["Destinations"] as! NSDictionary

// Retrieve the data service using the destinations dictionary and return it.
var dataService: ESPMContainer<OnlineODataProvider>? {
    guard let odataController = OnboardingSessionManager.shared.onboardingSession?.odataControllers[destinations["com.sap.edm.sampleservice.v2"] as! String] as? Comsapedmsampleservicev2OnlineODataController, let dataService = odataController.espmContainer else {
        AlertHelper.displayAlert(with: NSLocalizedString("OData service is not reachable, please onboard again.", comment: ""), error: nil, viewController: self)
        return nil
    }
    return dataService
}

private var products = [Product]()

Because the classification as well as the data loading will take some time, you should display a loading indicator to let the user know that your app is currently busy working on those tasks.

Every Xcode project generated by the iOS Assistant has a convenience protocol, making it easy for you to display and hide a SAPFioriLoadingIndicator.

Let the ProductClassificationTableViewController.swift class conform to the SAPFioriLoadingIndicator protocol:

class ProductClassificationTableViewController: UITableViewController, SAPFioriLoadingIndicator

The protocol wants you to add the FUILoadingIndicatorView as a property to your class.

Add the following line of code right above the dataService property:

var loadingIndicator: FUILoadingIndicatorView?

You will use an FUIObjectTableViewCell to display the products in the Table View, also you will need access to the data service.

Add the following lines of code to the viewDidLoad(_:) method:

tableView.estimatedRowHeight = 80
tableView.rowHeight = UITableView.automaticDimension

tableView.register(FUIObjectTableViewCell.self, forCellReuseIdentifier: FUIObjectTableViewCell.reuseIdentifier)

To classify an image, you can use the Vision framework to prepare the image for classification.

Import the Vision framework by adding the import statement right below the UIKit import above the class declaration:

import Vision

You will use a so called VNCoreMLRequest which will use an instance of the Core ML model for image classification.

Implement the following code right below the viewDidLoad(:), read the inline comments carefully:

lazy var classificationRequest: VNCoreMLRequest = {
    do {
        // Instantiate the Core ML model
        let model = try VNCoreMLModel(for: TechEd2019().model)

        // Create a VNCoreMLRequest passing in the model and starting the classification process in the completionHandler.
        let request = VNCoreMLRequest(model: model, completionHandler: { [weak self] request, error in
            self?.processClassifications(for: request, error: error)
        })

        // Crop and scale the image
        request.imageCropAndScaleOption = .centerCrop
        return request
    } catch {
        fatalError("Failed to load Vision ML model: \(error)")
    }
}()

Next you want to implement a method performing the requests.

Implement a method called updateClassifications(for:):

/// - Tag: PerformRequests
func updateClassifications(for image: UIImage) {

    // show the loading indicator
    self.showFioriLoadingIndicator("Finding similar products...")

    // make sure the orientation of the image is passed in the CGImagePropertyOrientation to set the orientation of the image
    let orientation = CGImagePropertyOrientation(image.imageOrientation)

    // Create a CIImage as needed by the model for classification. If that fails throw a fatalError.
    guard let ciImage = CIImage(image: image) else { fatalError("Unable to create \(CIImage.self) from \(image).") }

    // Dispatch to the Global queue to asynchronously perform the classification request.
    DispatchQueue.global(qos: .userInitiated).async {
        let handler = VNImageRequestHandler(ciImage: ciImage, orientation: orientation)
        do {
            try handler.perform([self.classificationRequest])
        } catch {
            /*
             This handler catches general image processing errors. The `classificationRequest`'s
             completion handler `processClassifications(_:error:)` catches errors specific
             to processing that request.
             */
            print("Failed to perform classification.\n\(error.localizedDescription)")
        }
    }
}

Right now the code above will cause compile time errors. You need an extension on CGImagePropertyOrientation to manually match the orientations of UIImage to the CGImagePropertyOrientation.

Create a new Swift class with the name CGImagePropertyOrientation+UIImageOrientation in the Project Navigator.

In that extension class, replace the code with the following lines:

import UIKit
import ImageIO

extension CGImagePropertyOrientation {
    init(_ orientation: UIImage.Orientation) {
        switch orientation {
        case .up: self = .up
        case .upMirrored: self = .upMirrored
        case .down: self = .down
        case .downMirrored: self = .downMirrored
        case .left: self = .left
        case .leftMirrored: self = .leftMirrored
        case .right: self = .right
        case .rightMirrored: self = .rightMirrored
        }
    }
}

Go back to the ProductClassificationTableViewController.swift class and implement a method processing the image classification. This method will also fetch the products according to the classification result.

Implement the following method directly below updateClassifications(for:) and read the inline comments carefully:

/// - Tag: ProcessClassifications
func processClassifications(for request: VNRequest, error: Error?) {

    // Use the main dispatch queue
    DispatchQueue.main.async {

        // Check if the results are nil and display the error in an Alert Dialogue
        guard let results = request.results else {
            self.logger.error("Unable to classify image.", error: error)
            AlertHelper.displayAlert(with: "Unable to classify image.", error: error, viewController: self)
            return
        }
        // The `results` will always be `VNClassificationObservation`s, as specified by the Core ML model in this project.
        let classifications = results as! [VNClassificationObservation]

        if classifications.isEmpty {
            AlertHelper.displayAlert(with: "Couldn't recognize the image", error: nil, viewController: self)
        } else {
            // Retrieve top classifications ranked by confidence.
            let topClassifications = classifications.prefix(2)
            let categoryNames = topClassifications.map { classification in
                return String(classification.identifier)
            }

            // Safe unwrap the first classification, because that will be the category with the highest confidence.
            guard let category = categoryNames.first else {
                AlertHelper.displayAlert(with: "Unable to identify product category", error: nil, viewController: self)
                self.logger.error("Something went wrong. Please check the classification code.")
                return
            }

            // Set the Navigation Bar's title to the classified category
            self.navigationItem.title = category

            // Define a DataQuery to only fetch the products matching the classified product category
            let query = DataQuery().filter(Product.category == category)

            // Fetch the products matching the defined query
            self.dataService?.fetchProducts(matching: query) { [weak self] result, error in
                if let error = error {
                    AlertHelper.displayAlert(with: "Failed to load list of products!", error: error, viewController: self!)
                    self?.logger.error("Failed to load list of products!", error: error)
                    return
                }

                // Hide the loading indicator
                self?.hideFioriLoadingIndicator()
                self?.products = result!

                // You will display the product images as well, for that reason create a new array containing the picture urls.
                self?.productImageURLs = result!.map { $0.pictureUrl ?? "" }
                self?.tableView.reloadData()
            }
        }
    }
}

Add the productImageURLs property right below the product array property above in the class:

private var productImageURLs = [String]()

The last step would be to call the updateClassifications(for:) method inside the viewDidLoad(_:) method as the last line of code:

updateClassifications(for: image)

Your viewDidLoad(_:) should now look like this:

override func viewDidLoad() {
    super.viewDidLoad()

    tableView.estimatedRowHeight = 80
    tableView.rowHeight = UITableView.automaticDimension

    tableView.register(FUIObjectTableViewCell.self, forCellReuseIdentifier: FUIObjectTableViewCell.reuseIdentifier)

    updateClassifications(for: image)
}

That’s all you need to do to classify an image with Vision and a pre-trained Core ML model.

Continue with the tutorial to implement the displaying of products in the Table View.

To display the products, you will implement the data source methods directly in the class like you have done before.

Replace the existing numberOfSections(in:) and the tableView(_:numberOfRowsInSection) methods with the following code:

// Return one section.
override func numberOfSections(in tableView: UITableView) -> Int {
    return 1
}

// The number of rows is dependant on the available products.
override func tableView(_ tableView: UITableView, numberOfRowsInSection section: Int) -> Int {
    return products.count
}

Next implement the tableView(_:cellForRowAt:) method and read the inline comments carefully:

override func tableView(_ tableView: UITableView, cellForRowAt indexPath: IndexPath) -> UITableViewCell {

    // Get the correct product to display
    let product = products[indexPath.row]

    // Dequeue the FUIObjectTableViewCell
    let cell = tableView.dequeueReusableCell(withIdentifier: FUIObjectTableViewCell.reuseIdentifier) as! FUIObjectTableViewCell

    // Set the properties of the Object Cell to the name and category name
    cell.headlineText = product.name ?? ""
    cell.subheadlineText = product.categoryName ?? ""

    // If there is a price available, format it using the NumberFormatter and set it to the footnoteText property of the Object Cell.
    if let price = product.price {
        let formatter = NumberFormatter()
        formatter.numberStyle = .currency
        let formattedPrice = formatter.string(for: price.intValue())

        cell.footnoteText = formattedPrice ?? ""
    }

    // Because you're using a lazy loading mechanism for displaying the product images to avoid lagging of the Table View, you have to set a placeholder image on the detailImageView property of the Object Cell.
    cell.detailImageView.image = FUIIconLibrary.system.imageLibrary

    // The data service will return the image url in the following format: /imgs/HT-2000.jpg
    // In order to build together the URL you have to define the base URL.
    // The Base URL is found in the Mobile Services API tab.
    let baseURL = "https://hcpms-d061070trial.hanatrial.ondemand.com/com.sap.edm.sampleservice.v2"
    let url = URL(string: baseURL.appending(productImageURLs[indexPath.row]))

    // Safe unwrap the URL, the code above could fail when the URL is not in the correct format, so you have to make sure it is safely unwrapped so you can react accordingly. You won't show the product image if the URL is nil.
    guard let unwrapped = url else {
        logger.info("URL for product image is nil. Returning cell without image.")
        return cell
    }
    // You will use an image cache to cache all already loaded images. If the image is already in the cache display it right out of the cache.
    if let img = imageCache[unwrapped.absoluteString] {
        cell.detailImageView.image = img
    }

    // If the image is not loaded yet, use the loadProductImageFrom(_:) method to load the image from the data service.
    else {
        // The image is not cached yet, so download it.
        loadProductImageFrom(unwrapped) { image in
            cell.detailImageView.image = image
        }
    }

    return cell
}

Before you will implement the loadProductImageFrom(_:) method, you have to define an image cache. As a cache you will simply use a dictionary.

Add the following property definition to below the private var productImageURLs = [String]() property:

private var imageCache = [String:UIImage]()

Next implement the loadProductImageFrom(_:) method to actually fetch the product images.

Implement the method right below the viewDidLoad(_:) method and read the inline comments carefully:

private func loadImageFrom(_ url: URL, completionHandler: @escaping (_ image: UIImage) -> Void) {
    let appDelegate = UIApplication.shared.delegate as! AppDelegate
    if let sapURLSession = appDelegate.sessionManager.onboardingSession?.sapURLSession {
        sapURLSession.dataTask(with: url, completionHandler: { data, _, error in

            if let error = error {
                self.logger.error("Failed to load image!", error: error)
                return
            }

            if let image = UIImage(data: data!) {
                // safe image in image cache
                self.imageCache[url.absoluteString] = image
                DispatchQueue.main.async { completionHandler(image) }
            }
        }).resume()
    }
}