首页 > 解决方案 > 将一个 VNCoreMLFeatureValueObservation 结果(3D 双精度数组)转换为多个 UIImage

问题描述

我有一个 coreml 模型,它在运行后返回一个 VNCoreMLFeatureValueObservation 对象,其中包含 1“MultiArray:Double 10 x IMG_SIZE x IMG_SIZE array”

我如何将其转换为 10 个 UIImage,每个 UIImage 具有 IMG_SIZE x IMG_SIZE 尺寸,并且它们的值为灰度?

标签: iosswiftmachine-learningcoreml

解决方案


在窥探了一下之后,我发现我必须添加这些帮助函数:

https://github.com/hollance/CoreMLHelpers到我的 Xcode 项目。并来自 MultiArray 初始化问题:https ://stackoverflow.com/a/44462908/403403

然后我拼凑了这个解决方案:

 let request = VNCoreMLRequest(model: model) { (request, error) in
            guard let results = request.results as? [VNCoreMLFeatureValueObservation] else {
                fatalError("Model failed to process image")
            }

            let obs : VNCoreMLFeatureValueObservation = (results.first)!
            let m: MLMultiArray = obs.featureValue.multiArrayValue!
            var mArrays = [MLMultiArray]()


            for i in 0..<10 {
                let start = i*(IMG_SIZE*IMG_SIZE) 
                guard let tmp : MLMultiArray = try? MLMultiArray(shape:[768,768], dataType:MLMultiArrayDataType.double) else {
                    fatalError("Unexpected runtime error. MLMultiArray")
                }
                for n in 0..<(IMG_SIZE*IMG_SIZE) {
                    tmp[n] = m[start+n]
                }
                mArrays.append(tmp)
            }



            self.imagePred0.image = mArrays[0].image(offset: 0, scale: 255)!
            self.imagePred1.image = mArrays[1].image(offset: 0, scale: 255)!
            self.imagePred2.image = mArrays[2].image(offset: 0, scale: 255)!
            self.imagePred3.image = mArrays[3].image(offset: 0, scale: 255)!
            self.imagePred4.image = mArrays[4].image(offset: 0, scale: 255)!
            self.imagePred5.image = mArrays[5].image(offset: 0, scale: 255)!
            self.imagePred6.image = mArrays[6].image(offset: 0, scale: 255)!
            self.imagePred7.image = mArrays[7].image(offset: 0, scale: 255)!
            self.imagePred8.image = mArrays[8].image(offset: 0, scale: 255)!
            self.imagePred9.image = mArrays[9].image(offset: 0, scale: 255)!



        }

希望有一种更清洁的方式,但现在可以使用


推荐阅读