I was playing around with the Vision framework to detect text from images based on this article: https://www.hackingwithswift.com/example-code/vision/how-to-use-vnrecognizetextrequests-optical-character-recognition-to-detect-text-in-an-image

However I'm running into some issues. In this line:

let handler = VNImageRequestHandler(cgImage: img, options: [:])

XCode is telling me that I need to explicitly cast the img as cgimage like this instead

let handler = VNImageRequestHandler(cgImage: img as! CGImage, options: [:]) 

But after this I keep getting a crash at the same line:

Any ideas why this might happen? The image is in the app bundle, so this shouldn't be the issue, also casting should work fine...

I've tried a few alternative solutions but haven't found a working one yet. Currently it seems that casting it to cgImage, which wasn't part of the sample code, but which the compiler requires, is causing the issue.

If I put a breakpoint and look at the NSImage (not UIImage as it's a macOS project), then the image is shown correctly, however later when it's cast to cgImage, then it seems to break, but not sure what's causing this.

Any ideas are welcome :)

Hi.

I could get it running in a Playground with the following code:

DispatchQueue.global(qos: .userInitiated).async {
    guard let nsImg = NSImage(named: "testImage.png") else {
        fatalError("missing image!")
    }
    guard let cgImg = nsImg.cgImage(forProposedRect: nil, context: nil, hints: nil) else {
        fatalError("can't convert image")
    }

    let handler = VNImageRequestHandler(cgImage: cgImg, options: [:])
    try? handler.perform(requests)
}

Wow, a few people stuck on this problem... today even! I'm having trouble with the CVImageBuffer and CGImage initiliazers too, will try to convert to a CMBuffer next, but that doesn't look too easy. I'm using the lastest Xcode after a full update to both OS, iOS and Xcode. I however get a seg fault and no better descrition except that it's While cross-referencing conformance for 'VNRecognizedPointKey' (in module 'Vision'). Hope some solutions are out there soon!