Question

我正在尝试使用自定义模型将CameraX与MLKit集成。我面临的问题是尝试使用从ImageAnalysis回调中获取的ProxyImage并在MLKit中使用该数据。以前使用Camera2，我将操纵SurfaceView上的位图。我可以通过创建FirebaseVisionImage来获取位图，但我想避免在收到的每一帧上操纵位图。这非常昂贵。

这在logcat中可见：ML Kit has detected that you seem to pass camera frames to the detector as a Bitmap object. This is inefficient. Please use YUV_420_888 format for camera2 API or NV21 format for (legacy) camera API and directly pass down the byte array to ML Kit

我需要从ImageAnalysis中获取图像数据，并获取相机预览所捕获内容的中间部分（我们正在使用全屏相机预览），因为我们的自定义模型需要接受方形图像。这是我目前拥有的：

 val preview = Preview.Builder().apply {
        setTargetResolution(screenSize)
        setTargetRotation(viewFinder.display.rotation)
    }.build()
    preview.previewSurfaceProvider = viewFinder.previewSurfaceProvider

    val imageAnalysis = ImageAnalysis.Builder()
            .setBackpressureStrategy(ImageAnalysis.STRATEGY_KEEP_ONLY_LATEST)
            .setTargetAspectRatio(AspectRatio.RATIO_16_9)
            .build()

    imageAnalysis.setAnalyzer(executor,
            ImageAnalysis.Analyzer {imageProxy ->
                \\DO image manipulation here
            })

    val cameraSelector = CameraSelector.Builder()
            .requireLensFacing(CameraSelector.LENS_FACING_BACK)
            .build()

    cameraProviderFuture.addListener(Runnable {
        val cameraProvider = cameraProviderFuture.get()

        cameraProvider.bindToLifecycle(
                requireActivity(),
                cameraSelector,
                preview,
                imageAnalysis
        )
    }, executor)

将CameraX与MLKit自定义模型集成的问题

0 个答案: