Android ML Kit 无法标记图像

Question

所以我有一个 Android 应用程序，用户可以使用相机应用程序为鸟儿拍照，然后它会对鸟儿进行分类。我遵循了label images with a custom model for Android 的文档，但它不起作用。我的onActivityResult中有这段代码：

val source = ImageDecoder.createSource(this.contentResolver, Uri.fromFile(photoFile))
val birdBitmap= ImageDecoder.decodeBitmap(source)

val image = InputImage.fromBitmap(birdBitmap, 0)
val localModel = LocalModel.Builder().setAssetFilePath("model.tflite").build()
val customImageLabelerOptions = CustomImageLabelerOptions.Builder(localModel)
    .setConfidenceThreshold(0.5f)
    .setMaxResultCount(5)
    .build()
val imageLabeler =
    ImageLabeling.getClient(customImageLabelerOptions)
imageLabeler.process(image)
    .addOnSuccessListener { labels ->
        var highConf = -1.0f
        var highText = ""
        for (label in labels) {
            val text = label.text
            val confidence = label.confidence
            val index = label.index

            if(confidence > highConf) {
                highConf = confidence
                highText = text
            }
        }
        Log.d("PREDICTION", "$highText, $highConf")
    }
    .addOnFailureListener { e ->
        Log.d("FAIL", "$e")
    }

当用户拍照时，addOnFailureListener 被调用，这是我得到的错误：

com.google.mlkit.common.MlKitException: Internal error has occurred when executing ML Kit tasks
at com.google.mlkit.common.sdkinternal.ModelResource.zza(com.google.mlkit:common@@16.0.0:28)
at com.google.mlkit.common.sdkinternal.zzn.call(Unknown Source:6)
at com.google.mlkit.common.sdkinternal.zzm.run(com.google.mlkit:common@@16.0.0:5)
at com.google.mlkit.common.sdkinternal.zzq.run(com.google.mlkit:common@@16.0.0:3)
at com.google.mlkit.common.sdkinternal.MlKitThreadPool.zzd(com.google.mlkit:common@@16.0.0:24)
at com.google.mlkit.common.sdkinternal.MlKitThreadPool.zza(com.google.mlkit:common@@16.0.0:30)
at com.google.mlkit.common.sdkinternal.zzj.run(Unknown Source:2)
At java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1167)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:641)
at java.lang.Thread.run(Thread.java:764)
Caused by: java.lang.IllegalArgumentException: Unsupported bitmap config HARDWARE
at com.google.android.libraries.vision.visionkit.pipeline.zzbb.zza(com.google.mlkit:vision-internal-vkp@@17.0.0:56)
at com.google.mlkit.vision.vkp.PipelineManager.process(com.google.mlkit:vision-internal-vkp@@17.0.0:150)
at com.google.mlkit.vision.label.custom.internal.zzd.zza(com.google.mlkit:image-labeling-custom@@16.1.0:19)
at com.google.mlkit.vision.label.custom.internal.zzd.run(com.google.mlkit:image-labeling-custom@@16.1.0:112)
at com.google.mlkit.vision.common.internal.MobileVisionBase.zza(com.google.mlkit:vision-common@@16.0.0:23)
at com.google.mlkit.vision.common.internal.zzb.call(Unknown Source:4)
at com.google.mlkit.common.sdkinternal.ModelResource.zza(com.google.mlkit:common@@16.0.0:26)

谁能告诉我如何解决这个问题？

Answer 1

查看堆栈跟踪，输入位图似乎是直接从相机获取的，并且驻留在内存中 (Bitmap.Config.HARDWARE)。 ML Kit 只支持 ARGB_8888 格式的位图，请尝试：

val newBitmap = myBitmap.copy(Bitmap.Config.ARGB_8888, myBitmap.isMutable())

当我们开发这个时，我们考虑的是保存在设备上的文件并将其加载为 ARGB_8888，而不是从设备摄像头加载。我将提交一个增强请求，看看我们是否可以涵盖这个用例。感谢举报！