【Android,Kotlin,TFLite】移动设备集成深度学习轻模型TFlite(图像分类篇)

Why i create it?

为了创建一个易用且易于集成的TFlite加载库, 所以TFLiteLoader应运而生

• 开源Github项目地址 TFLiteLoader

集成 ImageClassifier

依赖

allprojects {
	repositories {
		...
		maven { url 'https://jitpack.io' }
	}
}

dependencies {
	implementation 'com.github.mozhimen.TFLiteLoader:imageclassifier:1.0.2'
}

接入

1. 全局声明

private lateinit var _tFLiteImageClassifier: TFLiteImageClassifier

2. 在onCreate中进行初始化

_tFLiteImageClassifier = TFLiteImageClassifier.create("health_model.tflite", resultSize = 3)

3. 分类图片

val objList = _tFLiteImageClassifier.classify([你的bitmap], 0)

• 对返回数据的处理示例, 可以pull代码参考demo

val objList = _tFLiteImageClassifier.classify(rotateBitmap, 0)
Log.d(TAG, "analyze: $objList")
runOnUiThread {
	if (objList.isEmpty()) [email protected]
	objList.forEachIndexed { index, _ ->
    	_stringBuilder.append("${objList[index].title}: ${objList[index].confidence}").append(" ")
	}
	vb.imageClassifierRes.text = _stringBuilder.toString()
	_stringBuilder.clear()
}

对于返回值的说明

• List{Recognition}

data class Recognition(
    /**
     * 已识别事物的唯一标识符。特定于类，而不是实例
     * A unique identifier for what has been recognized. Specific to the class, not the instance of
     * the object.
     */
    val id: String?,
    /**
     * 显示名称以进行识别
     * Display name for the recognition.
     */
    val title: String?,
    /**
     * 这是一个相对于其他识别程度的可分类分数。越高越好
     * A sortable score for how good the recognition is relative to others. Higher should be better.
     */
    val confidence: Float?,
    /**
     * 源图像中可选的位置，用于识别对象的位置, 图像分类中不返回obj的位置
     * Optional location within the source image for the location of the recognized object.
     */
    private var location: RectF?
) {

    fun getLocation(): RectF {
        return RectF(location)
    }

    fun setLocation(location: RectF?) {
        this.location = location
    }

    override fun toString(): String {
        var resultString = ""
        if (id != null) {
            resultString += "[$id] "
        }
        if (title != null) {
            resultString += "$title "
        }
        if (confidence != null) {
            resultString += String.format("(%.1f%%) ", confidence * 100.0f)
        }
        if (location != null) {
            resultString += location.toString() + " "
        }
        return resultString.trim { it <= ' ' }
    }
}

完整demo代码

@PermissionKAnnor(permissions = [Manifest.permission.CAMERA])
class ImageClassifierActivity : BaseKActivity<ActivityImageClassifierBinding, BaseKViewModel>(R.layout.activity_image_classifier) {
    private lateinit var _tFLiteImageClassifier: TFLiteImageClassifier
//    private lateinit var _tFLiteLabelImageClassifier: TFLiteLabelImageClassifier
//    private lateinit var _tFImageClassifier: TFImageClassifier

    override fun initData(savedInstanceState: Bundle?) {
        PermissionK.initPermissions(this) {
            if (it) {
                initView(savedInstanceState)
            } else {
                PermissionK.applySetting(this)
            }
        }
    }

    override fun initView(savedInstanceState: Bundle?) {
        initLiteLoader()
        initCamera()
    }

    private fun initLiteLoader() {
        _tFLiteImageClassifier = TFLiteImageClassifier.create("health_model.tflite", resultSize = 3)
//        _tFLiteLabelImageClassifier = TFLiteLabelImageClassifier.create("?", "labels.txt", modelType = ModelType.QUANTIZED_EFFICIENTNET)
//        _tFImageClassifier = TFImageClassifier.create("output_graph.pb", "output_labels.txt", "input", 299, "output", 128f, 128f, 0.1f, 1)
    }

    private fun initCamera() {
        vb.imageClassifierPreview.initCamera(this, CameraSelector.DEFAULT_BACK_CAMERA)
        vb.imageClassifierPreview.setImageAnalyzer(_frameAnalyzer)
        vb.imageClassifierPreview.startCamera()
    }

    private val _frameAnalyzer: ImageAnalysis.Analyzer by lazy {
        object : ImageAnalysis.Analyzer {
            private val _reentrantLock = ReentrantLock()
            private val _stringBuilder = StringBuilder()

            @SuppressLint("UnsafeOptInUsageError", "SetTextI18n")
            override fun analyze(image: ImageProxy) {
                try {
                    _reentrantLock.lock()
                    val bitmap: Bitmap = if (image.format == ImageFormat.YUV_420_888) {
                        ImageConverter.yuv2Bitmap(image)!!
                    } else {
                        ImageConverter.jpeg2Bitmap(image)
                    }
                    val rotateBitmap = UtilKBitmap.rotateBitmap(bitmap, 90)

                    val objList = _tFLiteImageClassifier.classify(rotateBitmap, 0)
                    Log.d(TAG, "analyze: $objList")
                    runOnUiThread {
                        if (objList.isEmpty()) [email protected]
                        objList.forEachIndexed { index, _ ->
                            _stringBuilder.append("${objList[index].title}: ${objList[index].confidence}").append(" ")
                        }
                        vb.imageClassifierRes.text = _stringBuilder.toString()
                        _stringBuilder.clear()
                    }
                } finally {
                    _reentrantLock.unlock()
                }

                image.close()
            }
        }
    }
}

关于这里的框架代码, 可以参考我另一个开源框架库: SwiftKit ,不过因为还未完成, 没有完整的wiki, 过段时间推出

• 本示例代码所持引用:

implementation 'com.github.mozhimen.SwiftKit:basick:1.1.1'
implementation('com.github.mozhimen.SwiftKit:abilityk:1.1.1') {
	exclude group: 'com.mozhimen.abilityk.scank'
    exclude group: 'com.huawei.hms'
}
implementation 'com.github.mozhimen.SwiftKit:componentk:1.1.1'

综上所述: 集成是不是很简单, 那赶快试试吧