How to make .accurate request for CoreML Segmentation - ios

Question

Im working on a bakground removal app , where Im doing the task using this function , but this is not so accurate . I found that Im not requesting accurate output here . I need to implement two line in my function .
Two line :
var segmentationRequest = VNGeneratePersonSegmentationRequest()
segmentationRequest.qualityLevel = . accurate
My Function :
func removeBackground(image:UIImage) -> UIImage?{
let resizedImage = image.resized(to: CGSize(width: 513, height: 513))
if let pixelBuffer = resizedImage.pixelBuffer(width:
Int(resizedImage.size.width), height: Int(resizedImage.size.height)){
if let outputImage = (try? modelCore.prediction(image:
pixelBuffer))?.semanticPredictions.image(min: 0, max: 1, axes: (0,0,1)),
let outputCIImage = CIImage(image:outputImage){
if let maskImage = removeWhitePixels(image:outputCIImage),
let resizedCIImage = CIImage(image: resizedImage), let compositedImage =
composite(image: resizedCIImage, mask: maskImage){
return UIImage(ciImage: compositedImage).resized(to: CGSize(width:
image.size.width, height: image.size.height))
}
}
}
return nil
}

Answer 1

You Can use .accurate Quality level in your request like this
but if you will use .accurate it will take more time to give you proper output. and I geuss the quality can be used only in non-Custom Models .
func generatePhoto(Image: Image?) {
//generating instance of request
let request = VNGeneratePersonSegmentationRequest()
guard
//Convert Input image to CgImg
let originalImage = Image?.foregroundImage.cgImage else {
print("missing require image")
return
}
//Setting quality level
request.qualityLevel = .accurate
request.revision = VNGeneratePersonSegmentationRequestRevision1
request.outputPixelFormat = kCVPixelFormatType_OneComponent8
//Create reqhandler
let requestHandler = VNImageRequestHandler(cgImage: originalImage, options: [:])
do{
//Process request
try requestHandler.perform([request])
guard let mask = request.results?.first else {
print("error")
return
}
//Convert the pixelbuffer to maskImg to get maskImg
let maskImage = CIImage(cvPixelBuffer: mask.pixelBuffer)
print(maskImage)
self.mskImage = maskImage
let foreground = CIImage(cgImage: originalImage).oriented(.up)
guard let output = blendImages(foreground: foreground, mask: maskImage) else {
print("Error4")
return
}
//update photoOutput
print(output)
//Convert the CIImg to UIImg
if let photoResult = renderAsUIImage(output){
print("output",photoResult.size)
self.outputImage = photoResult
print(outputImage)
}
}
catch {
print("Error processing person segmentation request")
}
}