Vision face detection poor results on iPad

Update,

I gave today CIDetector a try instead of Vision and I see much better results. Here are the two versions:

Very poor results:

Vision

let request = VNDetectFaceRectanglesRequest()

let pixelBuffer: CVPixelBuffer = CMSampleBufferGetImageBuffer(sampleBuffer)!

let handle = VNImageRequestHandler(cvPixelBuffer: pixelBuffer)

try! handle.perform([request])

guard

let result = request.results, result.count > 0

else { return }

print(result.count)

let face = request.results!.first as! VNFaceObservation

Much better results, but not as good as Vision on Imac

CIDetector

let accuracy = [CIDetectorAccuracy: CIDetectorAccuracyHigh

self.faceDetector = CIDetector(ofType: CIDetectorTypeFace, context: nil, options: accuracy)!

guard let imageBuffer = CMSampleBufferGetImageBuffer(sampleBuffer) else { return }

let ciImage = CIImage(cvPixelBuffer: imageBuffer)

guard let faces = self.faceDetector.features(in: ciImage) as? [CIFaceFeature] else { return }

I´m not happy with this, but at the moment I must admit that Vision is not the best choice. Maybe there are some tricks to accelerate Vision, which I don´t know..

Any Ideas are welcome---

Thanks, Hartwig

I am also a bit disappointed by the vision face (rectangle) recognition results on iOS. One thing I noticed was that converting the image to grey scale and improving the contrast brought out a lot of additional faces.

I applied the CIFilters "CISharpenLuminance" and "CIPhotoEffectNoir" and played with them for a while. I can´t see any differences....