异常无法捕获 #235
Assignees
Labels
Comments
|
Hi @LYS86, can you provide an exported android studio project? Also can you let me know where you got your Yolo10/11 model from? |
|
关于该问题的讨论 yolov11 export to tflite does not work on Android GPU 模型导出，yolov8n，yolov9t不会触发崩溃。 |
|
Hi, @LYS86 Thank you for your understanding and patience. |
|
Hi @LYS86, I believe you are applying a GPU delegate or something like that? Can you show me the code and the context around where/how you apply the delegate? Preferably the more you share of your project the faster/less turn around it will take to dig into this. Thanks. |
|
@gaikwadrahul8 是的，同样的问题，yolo11的模型转换出了问题，一个有效的解决方法是不使用yolo的导出代码，自行转换模型 这份代码导出的模型会触发异常，导致崩溃，yolo11n_saved_model/yolo11n_float16.tflite from ultralytics import YOLO
model = YOLO("yolo11n.pt")
model.export(format="tflite")这里导出的模型不会触发异常，saved_model/yolo11n_float16.tflite，两个模型使用cpu无异常 onnx2tf -i yolo11n.onnximport android.util.Log
import org.tensorflow.lite.InterpreterApi
import org.tensorflow.lite.gpu.CompatibilityList
import org.tensorflow.lite.gpu.GpuDelegate
import java.nio.MappedByteBuffer
class test {
private var gpuDelegate: GpuDelegate? = null
protected lateinit var options: InterpreterApi.Options
private lateinit var interpreter: InterpreterApi
init {
CompatibilityList().use { compatibilityList ->
if (compatibilityList.isDelegateSupportedOnThisDevice) {
// val gpu=compatibilityList.bestOptionsForThisDevice
// gpuDelegate = GpuDelegate(gpu)
gpuDelegate=GpuDelegate()
}
}
}
private fun getOptions(isGPU: Boolean): InterpreterApi.Options {
val opts = InterpreterApi.Options().apply {
useNNAPI = true
}
if (!isGPU) return opts
if (gpuDelegate == null) {
Log.w("test", "GPU不支持")
return opts
}
return opts.apply { addDelegate(gpuDelegate) }
}
/**
* 加载模型
* @param path 模型文件路径,sd路径
* @param isGPU 是否使用GPU
*/
fun loadModel(path: String, isGPU: Boolean) {
options = getOptions(isGPU)
loadModel(path, options)
}
private fun loadModel(path: String, options: InterpreterApi.Options) {
try {
val modelBuffer: MappedByteBuffer = FileUtil.loadModel(path)
interpreter = InterpreterApi.create(modelBuffer, options)
} catch (e: Exception) {
throw RuntimeException("${e.message}")
}
}
}这份示例是否足够，是否需要更多信息。 |
|
Hi @LYS86, thanks for the extra information, I think that's good for now... I tried adjusting one of our examples to naively use a GPU delegate and it seems to be failing for a different reason, so there does seem to be a bug independent of the specific model: the example I used: git clone https://github.com/google-ai-edge/litert-samples.git
# open litert-samples/examples/image_classification/androidmodify android/app/build.gradle.kts (add two lines to dependencies): implementation("com.google.ai.edge.litert:litert-gpu:1.0.1")
implementation("com.google.ai.edge.litert:litert-gpu-api:1.0.1")In "ImageClassificationHelper.kt", after options is defined (~line 95) add: val gpuDelegate = GpuDelegate()
options.addDelegate(gpuDelegate)Then execute... this error occurs with the default models in that example. I should note I'm using an emulator (Pixel 8 Pro API 34-ext12 ARM). my logcat: |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
你好，我在使用yolo模型时遇到了一些问题，使用gpu时部署yolo10，yolo11模型，会崩溃，无法捕获异常
The text was updated successfully, but these errors were encountered: