mp_start crash

[westpilgrim63]

hi,i compiled a so lib on android using the build-aarch64-android.sh script you provided.
i use this so lib in my jni cpp sdk,but the app crashed on initialization.
if i commented out the mp_start sentence, the initialization runs successful .
how can i solve this problem?

[westpilgrim63]

my env is android rk3588.my mediapipe version is 0.8.11

[westpilgrim63]

2023-10-25 16:49:22.071 3460-3524/com.mholos.gesture A/libc: Fatal signal 11 (SIGSEGV), code 1 (SEGV_MAPERR), fault addr 0x0 in tid 3524 (mediapipe/3524), pid 3460 (.mholos.gesture)
2023-10-25 16:49:25.030 3537-3537/? E/audit: rate limit exceeded
2023-10-25 16:49:22.446 3537-3537/? E/libprotobuf-native: [libprotobuf ERROR external/protobuf/src/google/protobuf/wire_format_lite.cc:584] String field 'LogMessage.message' contains invalid UTF-8 data when serializing a protocol buffer. Use the 'bytes' type if you intend to send raw bytes.
2023-10-25 16:49:22.447 3537-3537/? A/DEBUG: *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** ***
2023-10-25 16:49:22.447 3537-3537/? A/DEBUG: Build fingerprint: 'Firefly/rk3588/rk3588:12/SP2A.220405.004/star10131103:userdebug/release-keys'
2023-10-25 16:49:22.447 3537-3537/? A/DEBUG: Revision: '0'
2023-10-25 16:49:22.447 3537-3537/? A/DEBUG: ABI: 'arm64'
2023-10-25 16:49:22.447 3537-3537/? A/DEBUG: Timestamp: 2023-10-25 08:49:22.203188525+0000
2023-10-25 16:49:22.447 3537-3537/? A/DEBUG: Process uptime: 0s
2023-10-25 16:49:22.447 3537-3537/? A/DEBUG: Cmdline: com.mholos.gesture
2023-10-25 16:49:22.447 3537-3537/? A/DEBUG: pid: 3460, tid: 3524, name: mediapipe/3524 >>> com.mholos.gesture <<<
2023-10-25 16:49:22.447 3537-3537/? A/DEBUG: uid: 10088
2023-10-25 16:49:22.447 3537-3537/? A/DEBUG: tagged_addr_ctrl: 0000000000000001
2023-10-25 16:49:22.447 3537-3537/? A/DEBUG: signal 11 (SIGSEGV), code 1 (SEGV_MAPERR), fault addr 0x0
2023-10-25 16:49:22.447 3537-3537/? A/DEBUG: Cause: null pointer dereference
2023-10-25 16:49:22.447 3537-3537/? A/DEBUG: x0 0000000000000000 x1 0000007a60f667cc x2 00000077d2e85620 x3 0000007799abf648
2023-10-25 16:49:22.447 3537-3537/? A/DEBUG: x4 0000000000000000 x5 0000007822e525aa x6 705f7269645f6568 x7 7a69732e5f687461
2023-10-25 16:49:22.447 3537-3537/? A/DEBUG: x8 0000000000000000 x9 000000000000000b x10 0000007fc62f6978 x11 fffffffffffffff5
2023-10-25 16:49:22.447 3537-3537/? A/DEBUG: x12 000000000000000b x13 0000007fc62f7fc5 x14 00000000fffffff5 x15 000000772212589c
2023-10-25 16:49:22.447 3537-3537/? A/DEBUG: x16 000000772237cef8 x17 0000007a60ed8880 x18 000000772a564000 x19 0000007799abf780
2023-10-25 16:49:22.447 3537-3537/? A/DEBUG: x20 0000007799abf7a0 x21 0000000000000000 x22 00000077d2e85620 x23 0000000000000020
2023-10-25 16:49:22.447 3537-3537/? A/DEBUG: x24 0000007799ac0000 x25 0000007799abf7a1 x26 0000000000000021 x27 00000077f2e81e00
2023-10-25 16:49:22.448 3537-3537/? A/DEBUG: x28 0000000000000040 x29 0000007799abf730
2023-10-25 16:49:22.448 3537-3537/? A/DEBUG: lr 00000077218f8428 sp 0000007799abf680 pc 0000007a60ed8890 pst 0000000080001000
2023-10-25 16:49:22.448 3537-3537/? A/DEBUG: backtrace:
2023-10-25 16:49:22.448 3537-3537/? A/DEBUG: #00 pc 0000000000049890 /apex/com.android.runtime/lib64/bionic/libc.so (__strlen_aarch64+16) (BuildId: cd7952cb40d1a2deca6420c2da7910be)
2023-10-25 16:49:22.448 3537-3537/? A/DEBUG: #1 pc 00000000005f2424 /data/app/~~I_GDm7pwwzw__wZXiF_YCw==/com.mholos.gesture-NM7V5nNeFvyFQk221AHutw==/base.apk!libmediapipe.so (mediapipe::PathToResourceAsFile(std::__ndk1::basic_string<char, std::__ndk1::char_traits, std::__ndk1::allocator > const&)+572)
2023-10-25 16:49:22.448 3537-3537/? A/DEBUG: #2 pc 0000000000588810 /data/app/~~I_GDm7pwwzw__wZXiF_YCw==/com.mholos.gesture-NM7V5nNeFvyFQk221AHutw==/base.apk!libmediapipe.so (mediapipe::LocalFileContentsCalculator::Open(mediapipe::CalculatorContext*)+376)
2023-10-25 16:49:22.448 3537-3537/? A/DEBUG: #3 pc 0000000000b17558 /data/app/~~I_GDm7pwwzw__wZXiF_YCw==/com.mholos.gesture-NM7V5nNeFvyFQk221AHutw==/base.apk!libmediapipe.so (mediapipe::CalculatorNode::OpenNode()+840)
2023-10-25 16:49:22.448 3537-3537/? A/DEBUG: #4 pc 0000000000b01224 /data/app/~~I_GDm7pwwzw__wZXiF_YCw==/com.mholos.gesture-NM7V5nNeFvyFQk221AHutw==/base.apk!libmediapipe.so (mediapipe::internal::SchedulerQueue::OpenCalculatorNode(mediapipe::CalculatorNode*)+276)
2023-10-25 16:49:22.448 3537-3537/? A/DEBUG: #5 pc 0000000000b00fc0 /data/app/~~I_GDm7pwwzw__wZXiF_YCw==/com.mholos.gesture-NM7V5nNeFvyFQk221AHutw==/base.apk!libmediapipe.so (mediapipe::internal::SchedulerQueue::RunNextTask()+220)
2023-10-25 16:49:22.448 3537-3537/? A/DEBUG: #6 pc 0000000000b38f70 /data/app/~~I_GDm7pwwzw__wZXiF_YCw==/com.mholos.gesture-NM7V5nNeFvyFQk221AHutw==/base.apk!libmediapipe.so (mediapipe::ThreadPool::RunWorker()+380)
2023-10-25 16:49:22.448 3537-3537/? A/DEBUG: #7 pc 0000000000b38bc0 /data/app/~~I_GDm7pwwzw__wZXiF_YCw==/com.mholos.gesture-NM7V5nNeFvyFQk221AHutw==/base.apk!libmediapipe.so (mediapipe::ThreadPool::WorkerThread::ThreadBody(void*)+1380)
2023-10-25 16:49:22.448 3537-3537/? A/DEBUG: #8 pc 00000000000b1810 /apex/com.android.runtime/lib64/bionic/libc.so (__pthread_start(void*)+264) (BuildId: cd7952cb40d1a2deca6420c2da7910be)
2023-10-25 16:49:22.448 3537-3537/? A/DEBUG: #9 pc 00000000000512f0 /apex/com.android.runtime/lib64/bionic/libc.so (__start_thread+64) (BuildId: cd7952cb40d1a2deca6420c2da7910be)

[westpilgrim63]

i see that you use the same build file on linux ,windows and android ,but should the android build file be a little different? i see in google-ai-edge/mediapipe#3361 that the graph is different in mobile android devices from pcs

[chnoblouch]

Hi @westpilgrim63, thanks for the detailed report!

It looks like MediaPipe can't find some graph file when starting up. You probably need to copy the data directory to your Android assets and call mp_init_asset_manager before starting your graph.

I'm not sure if the build file is the problem here, we managed to get our app running on Android using this one.

[westpilgrim63]

@

Hi @westpilgrim63, thanks for the detailed report!

It looks like MediaPipe can't find some graph file when starting up. You probably need to copy the data directory to your Android assets and call mp_init_asset_manager before starting your graph.

I'm not sure if the build file is the problem here, we managed to get our app running on Android using this one.

@chnoblouch do you have some code example used in android jni? i copy the initialization code from example to my jni so lib, but i guess maybe the init code in pc and in jni is a little different?

[chnoblouch]

Unfortunately, I haven't created a standalone Android example yet, but we did something like this in our app:

public class ExampleActivity extends Activity {
    @Override
    protected void onCreate(Bundle savedInstanceState) {
        super.onCreate(savedInstanceState);
        initMediaPipe(this, getCacheDir().getAbsolutePath());
    }
	
    private static native void initMediaPipe(Context androidContext, String cacheDirPath);
}

extern "C" JNIEXPORT void JNICALL
Java_com_example_ExampleActivity_initMediaPipe(JNIEnv* env, jclass obj, jobject assetManager, jstring cacheDirPath) {
    mp_init_asset_manager(env, assetManager, cacheDirPath);
	
    const char *path = "<path to mediapipe data>/modules/hand_landmark/hand_landmark_tracking_cpu.binarypb";
    mp_instance_builder *builder = mp_create_instance_builder(path, "image");

    mp_add_option_float(builder, "palmdetectioncpu__TensorsToDetectionsCalculator", "min_score_thresh", 0.6);
    mp_add_option_double(builder, "handlandmarkcpu__ThresholdingCalculator", "threshold", 0.2);
    mp_add_side_packet(builder, "num_hands", mp_create_packet_int(2));
    mp_add_side_packet(builder, "model_complexity", mp_create_packet_int(1));
    mp_add_side_packet(builder, "use_prev_landmarks", mp_create_packet_bool(true));

    mp_instance *instance = mp_create_instance(builder);
    mp_poller *landmarks_poller = mp_create_poller(instance, "multi_hand_landmarks");
	
    // Do something with the graph
}

The initialization code is very similar to the desktop example, there is just a new call to mp_init_asset_manager.

The problem on Android is loading the files. The graph (hand_landmark_tracking_cpu.binarypb) needs to be somewhere on the filesystem (not in the assets) so it can be loaded. The machine learning models (hand_landmark_full.tflite, palm_detection_full.tflite) and handedness.txt need to be in the root asset directory.

We could probably simplify this by adding a function to load graphs from the assets.

[westpilgrim63]

Unfortunately, I haven't created a standalone Android example yet, but we did something like this in our app:

public class ExampleActivity extends Activity {
    @Override
    protected void onCreate(Bundle savedInstanceState) {
        super.onCreate(savedInstanceState);
        initMediaPipe(this, getCacheDir().getAbsolutePath());
    }
	
    private static native void initMediaPipe(Context androidContext, String cacheDirPath);
}

extern "C" JNIEXPORT void JNICALL
Java_com_example_ExampleActivity_initMediaPipe(JNIEnv* env, jclass obj, jobject assetManager, jstring cacheDirPath) {
    mp_init_asset_manager(env, assetManager, cacheDirPath);
	
    const char *path = "<path to mediapipe data>/modules/hand_landmark/hand_landmark_tracking_cpu.binarypb";
    mp_instance_builder *builder = mp_create_instance_builder(path, "image");

    mp_add_option_float(builder, "palmdetectioncpu__TensorsToDetectionsCalculator", "min_score_thresh", 0.6);
    mp_add_option_double(builder, "handlandmarkcpu__ThresholdingCalculator", "threshold", 0.2);
    mp_add_side_packet(builder, "num_hands", mp_create_packet_int(2));
    mp_add_side_packet(builder, "model_complexity", mp_create_packet_int(1));
    mp_add_side_packet(builder, "use_prev_landmarks", mp_create_packet_bool(true));

    mp_instance *instance = mp_create_instance(builder);
    mp_poller *landmarks_poller = mp_create_poller(instance, "multi_hand_landmarks");
	
    // Do something with the graph
}

The initialization code is very similar to the desktop example, there is just a new call to mp_init_asset_manager.

The problem on Android is how to load the files. The graph (hand_landmark_tracking_cpu.binarypb) needs to be somewhere on the filesystem (not in the assets) so it can be loaded. The machine learning models (hand_landmark_full.tflite, palm_detection_full.tflite) and handedness.txt need to be in the root asset directory.

We could probably simplify this by adding a function to load graphs from the assets.

your timly reply help me a lot.thanks very much.i will try it tomorrow

[westpilgrim63]

@chnoblouch hi , i run init with success. but when the program reach mp_process ,the app crashed again.
my code is :
mp_image image;
image.data = frame.data;
image.width = frame.cols;
image.height = frame.rows;
image.format = mp_image_format_srgb;

    // Wrap the image in a packet and process it.
    CHECK_MP_RESULT(mp_process(instance, mp_create_packet_image(image)))

    // Wait until the image has been processed.
    CHECK_MP_RESULT(mp_wait_until_idle(instance))

the error log is :
2023-10-27 10:39:31.290 5257-5476/com.mholos.gesture E/cv::error(): OpenCV(4.8.0) Error: Insufficient memory (Failed to allocate 409668290592 bytes) in OutOfMemoryError, file /home/ci/opencv/modules/core/src/alloc.cpp, line 73

2023-10-27 10:39:31.292 5257-5476/com.mholos.gesture E/libc++abi: terminating with uncaught exception of type cv::Exception: OpenCV(4.8.0) /home/ci/opencv/modules/core/src/alloc.cpp:73: error: (-4:Insufficient memory) Failed to allocate 409668290592 bytes in function 'OutOfMemoryError'

--------- beginning of crash

2023-10-27 10:39:31.292 5257-5476/com.mholos.gesture A/libc: Fatal signal 6 (SIGABRT), code -1 (SI_QUEUE) in tid 5476 (mediapipe_gl_ru), pid 5257 (.mholos.gesture)

2023-10-27 10:39:34.734 5503-5503/? E/audit: rate limit exceeded

2023-10-27 10:39:31.670 5503-5503/? E/libprotobuf-native: [libprotobuf ERROR external/protobuf/src/google/protobuf/wire_format_lite.cc:584] String field 'LogMessage.message' contains invalid UTF-8 data when serializing a protocol buffer. Use the 'bytes' type if you intend to send raw bytes.

2023-10-27 10:39:31.670 5503-5503/? A/DEBUG: *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** ***

2023-10-27 10:39:31.670 5503-5503/? A/DEBUG: Build fingerprint: 'Firefly/rk3588/rk3588:12/SP2A.220405.004/star10131103:userdebug/release-keys'

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: Revision: '0'

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: ABI: 'arm64'

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: Timestamp: 2023-10-27 02:39:31.418693309+0000

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: Process uptime: 1s

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: Cmdline: com.mholos.gesture

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: pid: 5257, tid: 5476, name: mediapipe_gl_ru >>> com.mholos.gesture <<<

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: uid: 10106

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: tagged_addr_ctrl: 0000000000000001

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: signal 6 (SIGABRT), code -1 (SI_QUEUE), fault addr --------

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: Abort message: 'terminating with uncaught exception of type cv::Exception: OpenCV(4.8.0) /home/ci/opencv/modules/core/src/alloc.cpp:73: error: (-4:Insufficient memory) Failed to allocate 409668290592 bytes in function 'OutOfMemoryError'

'

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: x0 0000000000000000 x1 0000000000001564 x2 0000000000000006 x3 0000007943447780

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: x4 736f646277641f73 x5 736f646277641f73 x6 736f646277641f73 x7 7f7f7f7f7f7f7f7f

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: x8 00000000000000f0 x9 9e580233dab32ca7 x10 0000000000000000 x11 ffffff80fffffbdf

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: x12 0000000000000001 x13 000000007fffffff x14 00000000000fb768 x15 000000099c5975d9

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: x16 0000007c08b16050 x17 0000007c08af2db0 x18 0000000000000004 x19 0000000000001489

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: x20 0000000000001564 x21 00000000ffffffff x22 ffffff80ffffffc8 x23 00000079434479d0

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: x24 00000079434478b0 x25 00000079434478f0 x26 0000007943449000 x27 0000000000000000

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: x28 0000007943448568 x29 0000007943447800

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: lr 0000007c08aa5aa0 sp 0000007943447760 pc 0000007c08aa5acc pst 0000000000001000

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: backtrace:

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: #00 pc 000000000004facc /apex/com.android.runtime/lib64/bionic/libc.so (abort+164) (BuildId: cd7952cb40d1a2deca6420c2da7910be)

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: #1 pc 0000000000e8d23c /data/app/~~k6mnjF_JSdnexiLArM0muQ==/com.mholos.gesture-dd05Su4kGmPwIfZqRHY1gA==/base.apk!libmediapipe.so

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: #2 pc 0000000000e8d394 /data/app/~~k6mnjF_JSdnexiLArM0muQ==/com.mholos.gesture-dd05Su4kGmPwIfZqRHY1gA==/base.apk!libmediapipe.so

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: #3 pc 0000000000e8a298 /data/app/~~k6mnjF_JSdnexiLArM0muQ==/com.mholos.gesture-dd05Su4kGmPwIfZqRHY1gA==/base.apk!libmediapipe.so

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: #4 pc 0000000000e89ba0 /data/app/~~k6mnjF_JSdnexiLArM0muQ==/com.mholos.gesture-dd05Su4kGmPwIfZqRHY1gA==/base.apk!libmediapipe.so (__cxa_rethrow+196)

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: #5 pc 00000000000913ac /data/app/~~k6mnjF_JSdnexiLArM0muQ==/com.mholos.gesture-dd05Su4kGmPwIfZqRHY1gA==/base.apk!libobsensor_jni.so (cv::Mat::create(int, int const*, int)+1356) (BuildId: 85f155b418641893c482b7519cf36996311dfb60)

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: #6 pc 00000000000920e0 /data/app/~~k6mnjF_JSdnexiLArM0muQ==/com.mholos.gesture-dd05Su4kGmPwIfZqRHY1gA==/base.apk!libobsensor_jni.so (cv::Mat::create(cv::Size_, int)+108) (BuildId: 85f155b418641893c482b7519cf36996311dfb60)

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: #7 pc 00000000000c3754 /data/app/~~k6mnjF_JSdnexiLArM0muQ==/com.mholos.gesture-dd05Su4kGmPwIfZqRHY1gA==/base.apk!libobsensor_jni.so (cv::OutputArray::create(cv::Size, int, int, bool, cv::_OutputArray::DepthMask) const+136) (BuildId: 85f155b418641893c482b7519cf36996311dfb60)

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: #8 pc 000000000056d22c /data/app/~~k6mnjF_JSdnexiLArM0muQ==/com.mholos.gesture-dd05Su4kGmPwIfZqRHY1gA==/base.apk!libopencv_java4.so (cv::warpPerspective(cv::_InputArray const&, cv::OutputArray const&, cv::InputArray const&, cv::Size, int, int, cv::Scalar const&)+528) (BuildId: a4de418c8c6a4390de7780366c357d03726e92f8)

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: #9 pc 0000000000610094 /data/app/~~k6mnjF_JSdnexiLArM0muQ==/com.mholos.gesture-dd05Su4kGmPwIfZqRHY1gA==/base.apk!libmediapipe.so

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: #10 pc 000000000060a1e0 /data/app/~~k6mnjF_JSdnexiLArM0muQ==/com.mholos.gesture-dd05Su4kGmPwIfZqRHY1gA==/base.apk!libmediapipe.so (mediapipe::api2::ImageToTensorCalculator::Process(mediapipe::CalculatorContext*)+2616)

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: #11 pc 0000000000bb0470 /data/app/~~k6mnjF_JSdnexiLArM0muQ==/com.mholos.gesture-dd05Su4kGmPwIfZqRHY1gA==/base.apk!libmediapipe.so (mediapipe::CalculatorNode::ProcessNode(mediapipe::CalculatorContext*)+1004)

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: #12 pc 0000000000b8c6a8 /data/app/~~k6mnjF_JSdnexiLArM0muQ==/com.mholos.gesture-dd05Su4kGmPwIfZqRHY1gA==/base.apk!libmediapipe.so (mediapipe::internal::SchedulerQueue::RunCalculatorNode(mediapipe::CalculatorNode*, mediapipe::CalculatorContext*)+304)

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: #13 pc 0000000000b8c14c /data/app/~~k6mnjF_JSdnexiLArM0muQ==/com.mholos.gesture-dd05Su4kGmPwIfZqRHY1gA==/base.apk!libmediapipe.so (mediapipe::internal::SchedulerQueue::RunNextTask()+232)

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: #14 pc 0000000000ba33b4 /data/app/~~k6mnjF_JSdnexiLArM0muQ==/com.mholos.gesture-dd05Su4kGmPwIfZqRHY1gA==/base.apk!libmediapipe.so

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: #15 pc 0000000000b9dd3c /data/app/~~k6mnjF_JSdnexiLArM0muQ==/com.mholos.gesture-dd05Su4kGmPwIfZqRHY1gA==/base.apk!libmediapipe.so (mediapipe::GlContext::DedicatedThread::ThreadBody()+160)

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: #16 pc 0000000000b9d76c /data/app/~~k6mnjF_JSdnexiLArM0muQ==/com.mholos.gesture-dd05Su4kGmPwIfZqRHY1gA==/base.apk!libmediapipe.so (mediapipe::GlContext::DedicatedThread::ThreadBody(void*)+8)

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: #17 pc 00000000000b1810 /apex/com.android.runtime/lib64/bionic/libc.so (__pthread_start(void*)+264) (BuildId: cd7952cb40d1a2deca6420c2da7910be)

2023-10-27 10:39:31.671 5503-5503/? A/DEBUG: #18 pc 00000000000512f0 /apex/com.android.runtime/lib64/bionic/libc.so (__start_thread+64) (BuildId: cd7952cb40d1a2deca6420c2da7910be)

could you please help me one last time？i am deeply grateful!!!!!

[westpilgrim63]

solved. the reason is i compile mediapipe with opencv 4.5.0. but i use opencv 4.8.0 in my jni code. i replace 4.5.0 with 4.8.0 .and recompile mp .now i can run mp on android rk3588 successfully.