FirebaseVisionImage.fromMediaImage() 为什么会导致 OutOfMemoryError？

Question

CameraX是构建的，analyze()方法被调用并传入一张图片，然后使用close()方法关闭（删除）该图片。从这张图片创建FirebaseVisionImage并传递进行处理（文本识别）。代码示例和代码实验室存在差异，并且不实现与CameraX或使用旧API版本的TextRecognition。

  override fun analyze(imageProxy: ImageProxy) {
    if (isValidText) {
        imageProxy.close()
        return
    }
    val mediaImage = imageProxy.image // requires annotation
    val degrees = imageProxy.imageInfo.rotationDegrees
    val rotation = rotationDegreesToFirebaseRotation(degrees)
    if (mediaImage != null) {
        runTextRecognition(mediaImage, rotation)  // line 44
    }
    imageProxy.close()
} 


private fun runTextRecognition(mediaImage: Image, rotation: Int) {
    // Create FirebaseVisionImage from frame
    val visionImage = FirebaseVisionImage.fromMediaImage(mediaImage, rotation) // line 64
    val recognizer = FirebaseVision.getInstance()
        .onDeviceTextRecognizer
    recognizer.processImage(visionImage)
        .addOnSuccessListener { texts ->
            processTextRecognitionResult(texts!!, recognizer)
            if (isValidText) {
                recognizer.close()
                return@addOnSuccessListener
            }
        }
        .addOnFailureListener { e -> // Task failed with an exception
            e.printStackTrace()
        }
}

在我的项目中，我正在使用这些依赖项。

def firebase_version = '24.0.2'
def camerax_version = '1.0.0-beta02'
implementation "com.google.firebase:firebase-ml-vision:$firebase_version"
implementation "androidx.camera:camera-camera2:$camerax_version"
implementation "androidx.camera:camera-view:1.0.0-alpha09"
implementation "androidx.camera:camera-lifecycle:${camerax_version}"

这就是我构建CameraX的方式。

 private fun bindPreview(cameraProvider: ProcessCameraProvider) {
    // Get screen metrics used to setup camera for full screen resolution
    val metrics = DisplayMetrics().also { viewFinder?.display?.getRealMetrics(it) }
    val screenAspectRatio = aspectRatio(metrics.widthPixels, metrics.heightPixels)
    val rotation = viewFinder?.display?.rotation
    // Set up the preview use case to display camera preview
    val preview = Preview.Builder()// Request aspect ratio but no resolution
        .setTargetAspectRatio(screenAspectRatio)
        // Set initial target rotation
        .setTargetRotation(rotation!!)
        .build()

    // Choose the camera by requiring a lens facing
    val cameraSelector = CameraSelector.Builder()
        .requireLensFacing(CameraSelector.LENS_FACING_BACK)
        .build()

    val executor = Executors.newSingleThreadExecutor()

    // Must unbind the use-cases before rebinding them
    cameraProvider.unbindAll()


    val imageAnalyzer = ImageAnalysis.Builder()
        // Request aspect ratio but no resolution
        .setTargetAspectRatio(screenAspectRatio)
        // Set initial target rotation, have to call this again if rotation changes
        // during the lifecycle of this use case
        .setTargetRotation(rotation)
        .setBackpressureStrategy(ImageAnalysis.STRATEGY_KEEP_ONLY_LATEST)
        .build()
    imageAnalyzer.setAnalyzer(executor, analyzer)

    var camera = cameraProvider.bindToLifecycle(viewFinder?.context as LifecycleOwner, cameraSelector, preview, imageAnalyzer)
    // Attach the viewfinder's surface provider to preview use case
    preview.setSurfaceProvider(viewFinder?.createSurfaceProvider(camera.cameraInfo))
}

- Sorel

Answer 1

我通过切换到mlkit解决了这个问题。

首先更新app/build.gradle文件，使用mlkit代替firebase：

 // Add ML Kit dependencies
 implementation 'com.google.android.gms:play-services-mlkit-text-recognition:16.1.0'

接下来更新分析器以使用InputImage：

@androidx.camera.core.ExperimentalGetImage
private class TextAnalyzer(private val listener: TextListener) : ImageAnalysis.Analyzer {

    override fun analyze(imageProxy: ImageProxy) {
        val mediaImage: Image = imageProxy.image ?: return
        val image = InputImage.fromMediaImage(mediaImage, imageProxy.imageInfo.rotationDegrees)

        runTextRecognition(image)

        imageProxy.close()
    }

然后更新runTextRecognition为：

private fun runTextRecognition(image: InputImage) {
    val recognizer = TextRecognition.getClient()
    recognizer.process(image)

        ...
}

应该就是这样了。

这里 是提供更多细节的代码实验室。

- rrao