iOS Question MLKit Selfie Segmentation

[walterf25]

I was able to use the MLKit subject segmentation code in an Android Application which successfully removes the entire background from an image leaving only the subject in the image.

I have been trying to do the same thing for B4i, but have not had much success, I saw this post here from Erel, but this seems to remove the subject rather than the background, I have been trying for about a week now to find a solution and have tried different inline objective c modificatons, but have not been able to achieve my goal.

Most of the modifications I have tried are suggestions from ChatGPT and code I've found in other places,

Here's one example

-(void (^)(id, id)) createBlock:(CVPixelBufferRef)pixelBuffer {
    void (^block)(id result, id error) = ^void (MLKSegmentationMask* mask, id error) {
        if (error != nil || mask == nil) {
            [B4I shared].lastError = error;
            [self.bi raiseUIEvent:nil event:@"process_result::" params:@[@(false), [NSNull null]]];
        } else {
            // Get mask dimensions and pixel buffer
            CVPixelBufferRef maskPixelBuffer = mask.buffer;
            CIImage *ciMaskImage = [CIImage imageWithCVPixelBuffer:maskPixelBuffer];
           
            // Ensure mask is in grayscale
            ciMaskImage = [ciMaskImage imageByApplyingFilter:@"CIColorMatrix"
                                   withInputParameters:@{
                                       @"inputRVector": [CIVector vectorWithX:0 Y:0 Z:0 W:1],
                                       @"inputGVector": [CIVector vectorWithX:0 Y:0 Z:0 W:1],
                                       @"inputBVector": [CIVector vectorWithX:0 Y:0 Z:0 W:1],
                                       @"inputAVector": [CIVector vectorWithX:0 Y:0 Z:0 W:1]
                                   }];
           
            CIContext *temporaryContext = [CIContext contextWithOptions:nil];
           
            // Convert the original pixel buffer to CIImage
            CIImage *originalImage = [CIImage imageWithCVPixelBuffer:pixelBuffer];

            // Ensure original image and mask have matching dimensions
            if (CVPixelBufferGetWidth(maskPixelBuffer) != CVPixelBufferGetWidth(pixelBuffer) ||
                CVPixelBufferGetHeight(maskPixelBuffer) != CVPixelBufferGetHeight(pixelBuffer)) {
                // Resize original image to match mask dimensions if necessary
                originalImage = [originalImage imageByApplyingFilter:@"CILanczosScaleTransform"
                                                withInputParameters:@{
                                                    kCIInputScaleKey: @(CVPixelBufferGetWidth(maskPixelBuffer) / CVPixelBufferGetWidth(pixelBuffer)),
                                                    kCIInputAspectRatioKey: @(1.0)
                                                }];
            }

            // Blend the mask and the original image
            CIImage *maskedImage = [originalImage imageByApplyingFilter:@"CIBlendWithAlphaMask"
                                    withInputParameters:@{
                                        kCIInputMaskImageKey: ciMaskImage
                                    }];
           
            // Convert the result to a UIImage
            CGImageRef videoImage = [temporaryContext createCGImage:maskedImage
                    fromRect:CGRectMake(0, 0,
                    CVPixelBufferGetWidth(maskPixelBuffer),
                    CVPixelBufferGetHeight(maskPixelBuffer))];
                   
            UIImage *uiImage = [UIImage imageWithCGImage:videoImage];
            CGImageRelease(videoImage);
           
            // Raise event with the processed image
            [self.bi raiseUIEvent:nil event:@"process_result::" params:@[@(true), uiImage]];
        }
    };
    return block;
}

This code returns an image that is about 95% blank with some colored streaks going across the image, so obviously it doesn't work, Has anyone been able to remove the background from an image leaving only the subject, I realize that the library being used in this case is selfie segmentation and not subject segmentation, it seems subject segmentation only works on Android? I can not find any documentation from MLKit for iOS only selfie segmentation which I realize may be totally different.

Still I would like to find a solution, if anyone has any ideas or thoughts please let me know.

Regards,
Walter