[英]How to convert camera image to use it in tflite
I have an android app in Kotlin which depends on CameraX to analyze images and pass them to tensorflow lite image detection model ( not classification ).我在 Kotlin 中有一个 android 应用程序,它依赖于 CameraX 来分析图像并将它们传递给 tensorflow 精简图像检测 model(不是分类)。 The problem is we get YUV or RGBA Output images and we need an argb bitmap to pass to tflite model to detect objects we have been struggling for weeks and we have a deadline soon help please.
问题是我们得到 YUV 或 RGBA Output 图像,我们需要一个 argb bitmap 传递给 tflite model 来检测我们已经苦苦挣扎了数周的物体,我们很快就有了截止日期,请帮忙。
ImageAnalysis.Analyzer {
private val supportedImages = listOf(ImageFormat.FLEX_RGBA_8888, ImageFormat.FLEX_RGB_888)
private var data = ""
@SuppressLint("UnsafeOptInUsageError")
override fun analyze(image: ImageProxy) {
if (image.format in supportedImages) {
val options = ObjectDetector.ObjectDetectorOptions.builder()
.setScoreThreshold(0.2F)
.setBaseOptions(BaseOptions.builder().build())
.build()
val objectDetector =
ObjectDetector.createFromFileAndOptions(
context, Model.createModel(context, "model.tflite").path, options
)
val bitmap = toBitmap(image)
val tfImage = TensorImage.fromBitmap(bitmap)
val results: List<Detection> = objectDetector.detect(tfImage)
for (result in results) {
val categories = result.categories
for (category in categories) {
val label = category.label
val boundingBox = result.boundingBox
data += "label : $label | bounding : $boundingBox"
}
}
listener(data)
image.close()
}
}
private fun toBitmap(image: ImageProxy): Bitmap {
val r = image.planes[0].buffer.int
val g = image.planes[1].buffer.int
val b = image.planes[2].buffer.int
val a = image.planes[3].buffer.int
var color = Color.argb(a, r, g, b)
val rect = Rect(0, 0, 1, 1)
color = ColorUtils.compositeColors(color, Color.WHITE)
val bitmap = Bitmap.createBitmap(rect.width(), rect.height(), Bitmap.Config.ARGB_8888)
val canvas = Canvas(bitmap)
val paint = Paint()
paint.color = color
canvas.drawRect(rect, paint)
savePic(bitmap = bitmap!!, display_name = Random(100000L).toString())
return bitmap
}
private fun savePic(bitmap: Bitmap, display_name: String): Boolean {
val mediaCollection = latest {
MediaStore.Images.Media.getContentUri(MediaStore.VOLUME_EXTERNAL_PRIMARY)
} ?: MediaStore.Images.Media.EXTERNAL_CONTENT_URI
val contentValues = ContentValues().apply {
put(MediaStore.Images.Media.DISPLAY_NAME, "${display_name}.jpg")
put(MediaStore.Images.Media.MIME_TYPE, "image/jpeg")
put(MediaStore.Images.Media.WIDTH, bitmap.width)
put(MediaStore.Images.Media.HEIGHT, bitmap.height)
}
return try {
context.contentResolver.insert(mediaCollection, contentValues)?.also { uri ->
context.contentResolver.openOutputStream(uri).use { outputStream ->
if (!bitmap.compress(Bitmap.CompressFormat.JPEG, 100, outputStream)) {
throw IOException("could not save bitmap!")
}
}
} ?: throw IOException("could not create media store entry")
return true
} catch (e: IOException) {
e.printStackTrace()
false
}
}
}```
Here is the function that converts YUV_420_888 imageproxy into argb format bitmap.这里是将YUV_420_888 imageproxy转成argb格式的function bitmap。
I know a late answer but providing for others.我知道一个迟到的答案,但为其他人提供。
fun imageToBitmap(image: Image): Bitmap {
val planes: Array<Image.Plane> = image.planes
val yBuffer: ByteBuffer = planes[0].buffer
val uBuffer: ByteBuffer = planes[1].buffer
val vBuffer: ByteBuffer = planes[2].buffer
val ySize = yBuffer.remaining()
val uSize = uBuffer.remaining()
val vSize = vBuffer.remaining()
val nv21 = ByteArray(ySize + uSize + vSize)
//U and V are swapped
yBuffer[nv21, 0, ySize]
vBuffer[nv21, ySize, vSize]
uBuffer[nv21, ySize + vSize, uSize]
val yuvImage = YuvImage(nv21, ImageFormat.NV21, image.width, image.height, null)
val out = ByteArrayOutputStream()
yuvImage.compressToJpeg(Rect(0, 0, yuvImage.width, yuvImage.height), 75, out)
val imageBytes: ByteArray = out.toByteArray()
return BitmapFactory.decodeByteArray(imageBytes, 0, imageBytes.size)
}
Provide imageproxy from cameraX analyser to this function and it will provide the bitmap.从 cameraX 分析器向这个 function 提供图像代理,它将提供 bitmap。
声明:本站的技术帖子网页,遵循CC BY-SA 4.0协议,如果您需要转载,请注明本站网址或者原文地址。任何问题请咨询:yoyou2525@163.com.