[mob][droid] Use custom dart plugin for face detection

2024-07-08 17:41:37 +05:30
parent 37519179a1
commit 273e5b88eb
21 changed files with 913 additions and 301 deletions
--- a/mobile/android/app/src/main/kotlin/io/ente/photos/MainActivity.kt
+++ b/mobile/android/app/src/main/kotlin/io/ente/photos/MainActivity.kt
@@ -3,11 +3,10 @@ package io.ente.photos
 import io.flutter.embedding.android.FlutterFragmentActivity
 import io.flutter.embedding.engine.FlutterEngine
 import io.flutter.plugins.GeneratedPluginRegistrant
-import io.ente.photos.ml.onnx.EnteOnnxFlutterPlugin

 class MainActivity : FlutterFragmentActivity() {
    override fun configureFlutterEngine(flutterEngine: FlutterEngine) {
        GeneratedPluginRegistrant.registerWith(flutterEngine)
-        flutterEngine.plugins.add(EnteOnnxFlutterPlugin())
+
    }
 }
--- a/mobile/android/app/src/main/kotlin/io/ente/photos/ml/onnx/EnteOnnx.kt
+++ b/mobile/android/app/src/main/kotlin/io/ente/photos/ml/onnx/EnteOnnx.kt
@@ -1,207 +0,0 @@
-package io.ente.photos.ml.onnx
-
-import android.content.Context
-import androidx.annotation.NonNull
-import ai.onnxruntime.*
-import io.flutter.embedding.engine.plugins.FlutterPlugin
-import io.flutter.plugin.common.MethodCall
-import io.flutter.plugin.common.MethodChannel
-import io.flutter.plugin.common.MethodChannel.MethodCallHandler
-import io.flutter.plugin.common.MethodChannel.Result
-import kotlinx.coroutines.*
-import java.nio.FloatBuffer
-import java.util.concurrent.ConcurrentHashMap
-import android.util.Log
-import java.io.File
-import java.util.concurrent.ConcurrentLinkedQueue
-
-object LongArrayPool {
-    private val pool = ConcurrentLinkedQueue<LongArray>()
-
-    fun get(size: Int): LongArray {
-        return pool.poll() ?: LongArray(size)
-    }
-
-    fun release(array: LongArray) {
-        pool.offer(array)
-    }
-}
-
-class EnteOnnxFlutterPlugin : FlutterPlugin, MethodCallHandler {
-    private var faceOrtEnv: OrtEnvironment = OrtEnvironment.getEnvironment()
-    private lateinit var channel: MethodChannel
-    private val scope = CoroutineScope(Dispatchers.IO + SupervisorJob())
-    private val sessionMap = ConcurrentHashMap<ModelType, ModelState>()
-    private lateinit var context: Context
-
-    companion object {
-        const val DEFAULT_SESSION_COUNT = 1
-        const val K_INPUT_WIDTH = 640
-        const val K_INPUT_HEIGHT = 640
-        const val K_NUM_CHANNELS = 3
-    }
-
-    enum class ModelType {
-        CLIP_TEXT, CLIP_VISUAL, YOLO_FACE, MOBILENET_FACE
-    }
-
-    data class ModelState(
-        var isInitialized: Boolean = false,
-        val sessionAddresses: ConcurrentHashMap<Int, OrtSession> = ConcurrentHashMap(),
-        // number of sessions that should have been created for given model
-        var sessionsCount: Int = DEFAULT_SESSION_COUNT
-    )
-
-    override fun onAttachedToEngine(@NonNull flutterPluginBinding: FlutterPlugin.FlutterPluginBinding) {
-        channel = MethodChannel(flutterPluginBinding.binaryMessenger, "ente_onnx_flutter_plugin")
-        channel.setMethodCallHandler(this)
-        context = flutterPluginBinding.applicationContext
-    }
-
-    override fun onDetachedFromEngine(@NonNull binding: FlutterPlugin.FlutterPluginBinding) {
-        channel.setMethodCallHandler(null)
-        releaseAllSessions()
-        scope.cancel()
-    }
-
-    override fun onMethodCall(@NonNull call: MethodCall, @NonNull result: Result) {
-        when (call.method) {
-            "init" -> {
-                val modelType = call.argument<String>("modelType") ?: run {
-                    result.error("INVALID_ARGUMENT", "Model type is missing", null)
-                    return
-                }
-                val modelPath = call.argument<String>("modelPath") ?: run {
-                    result.error("INVALID_ARGUMENT", "Model path is missing", null)
-                    return
-                }
-                val sessionsCount = call.argument<Int>("sessionsCount") ?: DEFAULT_SESSION_COUNT
-                init(ModelType.valueOf(modelType), modelPath, sessionsCount, result)
-            }
-            "release" -> {
-                val modelType = call.argument<String>("modelType") ?: run {
-                    result.error("INVALID_ARGUMENT", "Model type is missing", null)
-                    return
-                }
-                release(ModelType.valueOf(modelType), result)
-            }
-            "predict" -> {
-                val sessionAddress = call.argument<Int>("sessionAddress")
-                val inputData = call.argument<List<Double>>("inputData")
-                val modelType = call.argument<String>("modelType") ?: run {
-                    result.error("INVALID_ARGUMENT", "Model type is missing", null)
-                    return
-                }
-                if (sessionAddress == null || inputData == null) {
-                    result.error("INVALID_ARGUMENT", "Session address or input data is missing", null)
-                    return
-                }
-                val inputDataArray = inputData.map { it.toFloat() }.toFloatArray()
-                predict(ModelType.valueOf(modelType), sessionAddress, inputDataArray, result)
-            }
-            else -> result.notImplemented()
-        }
-    }
-
-    private fun readModelFile(modelPath: String): ByteArray {
-        return File(modelPath).readBytes()
-    }
-
-    private fun init(modelType: ModelType, modelPath: String, sessionsCount: Int, result: Result) {
-        scope.launch {
-            val modelState: ModelState
-            if (sessionMap.containsKey(modelType)) {
-                modelState = sessionMap[modelType]!!
-            } else {
-                modelState = ModelState()
-                sessionMap[modelType] = modelState
-            }
-            if (!modelState.isInitialized) {
-                for (i in 0 until sessionsCount) {
-                    val session = createSession(faceOrtEnv, modelPath)
-                    if (session != null) {
-                        modelState.sessionAddresses[i] = session
-                    }
-                }
-                modelState.isInitialized = true
-                modelState.sessionsCount = sessionsCount
-                withContext(Dispatchers.Main) {
-                    result.success(true)
-                }
-            } else {
-                withContext(Dispatchers.Main) {
-                    result.success(false)
-                }
-            }
-        }
-    }
-
-    private fun release(modelType: ModelType, result: Result) {
-        scope.launch {
-            val modelState = sessionMap[modelType]
-            modelState?.let {
-                it.sessionAddresses.forEach { entry: Map.Entry<Int, OrtSession> ->
-                    entry.value.close()
-                }
-                it.sessionAddresses.clear()
-                it.isInitialized = false
-            }
-            withContext(Dispatchers.Main) {
-                result.success(true)
-            }
-        }
-    }
-
-    private fun predict(modelType: ModelType, sessionAddress: Int, inputData: FloatArray, result: Result) {
-        scope.launch {
-            val modelState = sessionMap[modelType]
-            val session = modelState?.sessionAddresses?.get(sessionAddress)
-            if (session == null) {
-                withContext(Dispatchers.Main) {
-                    result.error("SESSION_NOT_FOUND", "Session not found for address: $sessionAddress", null)
-                }
-                return@launch
-            }
-
-            try {
-                val env = OrtEnvironment.getEnvironment()
-                val inputTensorShape = LongArrayPool.get(4).apply {
-                    this[0] = 1
-                    this[1] = K_NUM_CHANNELS.toLong()
-                    this[2] = K_INPUT_HEIGHT.toLong()
-                    this[3] = K_INPUT_WIDTH.toLong()
-                }
-                val inputTensor = OnnxTensor.createTensor(env, FloatBuffer.wrap(inputData), inputTensorShape)
-                val inputs = mapOf("input" to inputTensor)
-                val outputs = session.run(inputs)
-                Log.d("OnnxFlutterPlugin", "Output shape: ${outputs.size()}")
-
-                inputTensor.close()
-                outputs.close()
-                LongArrayPool.release(inputTensorShape)
-                withContext(Dispatchers.Main) {
-                    val dummyResult = listOf(0.1, 0.2) // Replace with actual result processing
-                    result.success(dummyResult)
-                }
-            } catch (e: OrtException) {
-                withContext(Dispatchers.Main) {
-                    result.error("PREDICTION_ERROR", "Error during prediction: ${e.message}", null)
-                }
-            }
-        }
-    }
-
-    private fun createSession(env: OrtEnvironment, modalPath: String): OrtSession? {
-        return env.createSession(modalPath, OrtSession.SessionOptions())
-    }
-
-    private fun releaseAllSessions() {
-        sessionMap.forEach { (_, modelState) ->
-            modelState.sessionAddresses.forEach { entry ->
-                entry.value.close()
-            }
-            modelState.sessionAddresses.clear()
-        }
-        sessionMap.clear()
-    }
-}
--- a/mobile/ios/Podfile.lock
+++ b/mobile/ios/Podfile.lock
@@ -466,7 +466,7 @@ SPEC CHECKSUMS:
  package_info_plus: 115f4ad11e0698c8c1c5d8a689390df880f47e85
  path_provider_foundation: 3784922295ac71e43754bd15e0653ccfd36a147c
  permission_handler_apple: 9878588469a2b0d0fc1e048d9f43605f92e6cec2
-  photo_manager: 4f6810b7dfc4feb03b461ac1a70dacf91fba7604
+  photo_manager: ff695c7a1dd5bc379974953a2b5c0a293f7c4c8a
  PromisesObjC: f5707f49cb48b9636751c5b2e7d227e43fba9f47
  receive_sharing_intent: 6837b01768e567fe8562182397bf43d63d8c6437
  screen_brightness_ios: 715ca807df953bf676d339f11464e438143ee625
--- a/mobile/lib/nativeplugins/onnx.dart
+++ b/mobile/lib/nativeplugins/onnx.dart
@@ -1,41 +0,0 @@
-import 'package:flutter/services.dart';
-
-class OnnxFlutterPlugin {
-  static const MethodChannel _channel =
-      MethodChannel('ente_onnx_flutter_plugin');
-
-  static Future<bool> init(
-    String modelType,
-    String modelPath, {
-    int sessionsCount = 1,
-  }) async {
-    final bool result = await _channel.invokeMethod('init', {
-      'modelType': modelType,
-      'modelPath': modelPath,
-      'sessionsCount': sessionsCount,
-    });
-    return result;
-  }
-
-  static Future<bool> release(String modelType) async {
-    final bool result =
-        await _channel.invokeMethod('release', {'modelType': modelType});
-    return result;
-  }
-
-  static Future<List<double>> predict(
-    List<double> inputData,
-    String modelType, {
-    int sessionAddress = 0,
-  }) async {
-    final List<dynamic> result = await _channel.invokeMethod(
-      'predict',
-      {
-        'sessionAddress': sessionAddress,
-        'inputData': inputData,
-        'modelType': modelType,
-      },
-    );
-    return result.cast<double>();
-  }
-}
--- a/mobile/lib/services/machine_learning/face_ml/face_detection/face_detection_service.dart
+++ b/mobile/lib/services/machine_learning/face_ml/face_detection/face_detection_service.dart
@@ -1,11 +1,12 @@
 import "dart:async";
 import "dart:developer" as dev show log;
-import "dart:io" show File;
-import 'dart:typed_data' show ByteData;
+import "dart:io" show File, Platform;
+import 'dart:typed_data' show ByteBuffer, ByteData, Float32List, Uint8List;
 import 'dart:ui' as ui show Image;

 import "package:computer/computer.dart";
 import 'package:logging/logging.dart';
+import "package:onnx_dart/onnx_dart.dart";
 import 'package:onnxruntime/onnxruntime.dart';
 import "package:photos/face/model/dimension.dart";
 import 'package:photos/services/machine_learning/face_ml/face_detection/detection.dart';
@@ -34,6 +35,7 @@ class FaceDetectionService {
  static const double kIouThreshold = 0.4;
  static const double kMinScoreSigmoidThreshold = 0.7;
  static const int kNumKeypoints = 5;
+  static bool useCustomPlugin = Platform.isAndroid;

  bool isInitialized = false;

@@ -49,19 +51,23 @@ class FaceDetectionService {
      final model =
          await RemoteAssetsService.instance.getAsset(modelRemotePath);
      final startTime = DateTime.now();
-      sessionAddress = await _computer.compute(
-        _loadModel,
-        param: {
-          "modelPath": model.path,
-        },
-      );
+      if (useCustomPlugin) {
+        final OnnxDart plugin = OnnxDart();
+        final bool? initResult = await plugin.init("YOLO_FACE", model.path);
+        isInitialized = initResult ?? false;
+      } else {
+        sessionAddress = await _computer.compute(
+          _loadModel,
+          param: {
+            "modelPath": model.path,
+          },
+        );
+        isInitialized = sessionAddress != -1;
+      }
      final endTime = DateTime.now();
      _logger.info(
        "Face detection model loaded, took: ${(endTime.millisecondsSinceEpoch - startTime.millisecondsSinceEpoch).toString()}ms",
      );
-      if (sessionAddress != -1) {
-        isInitialized = true;
-      }
    }
  }

@@ -80,7 +86,10 @@ class FaceDetectionService {
    ByteData imageByteData,
    int sessionAddress,
  ) async {
-    assert(sessionAddress != 0 && sessionAddress != -1);
+    assert(
+      !useCustomPlugin ? (sessionAddress != 0 && sessionAddress != -1) : true,
+      'sessionAddress should be valid',
+    );

    final stopwatch = Stopwatch()..start();

@@ -94,18 +103,6 @@ class FaceDetectionService {
      requiredHeight: kInputHeight,
      maintainAspectRatio: true,
    );
-
-    final inputShape = [
-      1,
-      3,
-      kInputHeight,
-      kInputWidth,
-    ];
-    final inputOrt = OrtValueTensor.createTensorWithDataList(
-      inputImageList,
-      inputShape,
-    );
-    final inputs = {'input': inputOrt};
    stopwatchPreprocessing.stop();
    dev.log(
      'Face detection image preprocessing is finished, in ${stopwatchPreprocessing.elapsedMilliseconds}ms',
@@ -117,39 +114,90 @@ class FaceDetectionService {

    // Run inference
    final stopwatchInterpreter = Stopwatch()..start();
-    List<OrtValue?>? outputs;
+
+    List<List<List<double>>>? nestedResults = [];
    try {
-      final runOptions = OrtRunOptions();
-      final session = OrtSession.fromAddress(sessionAddress);
-      outputs = session.run(runOptions, inputs);
-      // inputOrt.release();
-      // runOptions.release();
+      if (useCustomPlugin) {
+        nestedResults = await _runCustomPlugin(inputImageList);
+      } else {
+        nestedResults = _runJNIBasedPlugin(
+          sessionAddress,
+          inputImageList,
+        ); // [1, 25200, 16]
+      }
    } catch (e, s) {
-      _logger.severe('Error while running inference: $e \n $s');
+      dev.log('Error while running inference', error: e, stackTrace: s);
      throw YOLOFaceInterpreterRunException();
    }
    stopwatchInterpreter.stop();
-    _logger.info(
-      'interpreter.run is finished, in ${stopwatchInterpreter.elapsedMilliseconds} ms',
+    try {
+      _logger.info(
+        'interpreter.run is finished, in ${stopwatchInterpreter.elapsedMilliseconds} ms',
+      );
+
+      final relativeDetections =
+          _yoloPostProcessOutputs(nestedResults!, newSize);
+      stopwatch.stop();
+      _logger.info(
+        'predict() face detection executed in ${stopwatch.elapsedMilliseconds}ms',
+      );
+
+      return (relativeDetections, originalSize);
+    } catch (e, s) {
+      _logger.severe('Error while post processing', e, s);
+      rethrow;
+    }
+  }
+
+  static List<List<List<double>>>? _runJNIBasedPlugin(
+    int sessionAddress,
+    Float32List inputImageList,
+  ) {
+    final inputShape = [
+      1,
+      3,
+      kInputHeight,
+      kInputWidth,
+    ];
+    final inputOrt = OrtValueTensor.createTensorWithDataList(
+      inputImageList,
+      inputShape,
    );
+    final inputs = {'input': inputOrt};

-    final relativeDetections = _yoloPostProcessOutputs(outputs, newSize);
+    final runOptions = OrtRunOptions();
+    final session = OrtSession.fromAddress(sessionAddress);
+    final List<OrtValue?>? outputs = session.run(runOptions, inputs);
+    // inputOrt.release();
+    // runOptions.release();
+    return outputs?[0]?.value as List<List<List<double>>>; // [1, 25200, 16]
+  }

-    stopwatch.stop();
-    _logger.info(
-      'predict() face detection executed in ${stopwatch.elapsedMilliseconds}ms',
+  static Future<List<List<List<double>>>> _runCustomPlugin(
+    Float32List inputImageList,
+  ) async {
+    final OnnxDart plugin = OnnxDart();
+    final result = await plugin.predict(
+      inputImageList,
+      "YOLO_FACE",
    );
-
-    return (relativeDetections, originalSize);
+    final List<List<List<double>>> reconstructedTensor = [];
+    for (int i = 0; i < result.length; i += 25200 * 16) {
+      final List<List<double>> outerArray = [];
+      for (int j = 0; j < 25200; j++) {
+        final List<double> innerArray =
+            result.sublist(i + j * 16, i + (j + 1) * 16).cast<double>();
+        outerArray.add(innerArray);
+      }
+      reconstructedTensor.add(outerArray);
+    }
+    return reconstructedTensor;
  }

  static List<FaceDetectionRelative> _yoloPostProcessOutputs(
-    List<OrtValue?>? outputs,
+    List<List<List<double>>> nestedResults,
    Dimensions newSize,
  ) {
-    // // Get output tensors
-    final nestedResults =
-        outputs?[0]?.value as List<List<List<double>>>; // [1, 25200, 16]
    final firstResults = nestedResults[0]; // [25200, 16]

    // Filter output
@@ -160,11 +208,6 @@ class FaceDetectionService {
      results: firstResults,
    );

-    // Release outputs
-    // outputs?.forEach((element) {
-    //   element?.release();
-    // });
-
    // Account for the fact that the aspect ratio was maintained
    for (final faceDetection in relativeDetections) {
      faceDetection.correctForMaintainedAspectRatio(
--- a/mobile/lib/services/machine_learning/face_ml/face_ml_service.dart
+++ b/mobile/lib/services/machine_learning/face_ml/face_ml_service.dart
@@ -744,6 +744,7 @@ class FaceMlService {
      final sendPort = message[2] as SendPort;

      try {
+        await FaceDetectionService.instance.init();
        switch (function) {
          case FaceMlOperation.analyzeImage:
            final time = DateTime.now();
@@ -838,7 +839,6 @@ class FaceMlService {

    final String? filePath =
        await _getImagePathForML(enteFile, typeOfData: FileDataForML.fileData);
-
    if (filePath == null) {
      _logger.warning(
        "Failed to get any data for enteFile with uploadedFileID ${enteFile.uploadedFileID} since its file path is null",
--- a/mobile/plugins/onnx_dart/.metadata
+++ b/mobile/plugins/onnx_dart/.metadata
@@ -0,0 +1,30 @@
+# This file tracks properties of this Flutter project.
+# Used by Flutter tool to assess capabilities and perform upgrades etc.
+#
+# This file should be version controlled and should not be manually edited.
+
+version:
+  revision: "761747bfc538b5af34aa0d3fac380f1bc331ec49"
+  channel: "[user-branch]"
+
+project_type: plugin
+
+# Tracks metadata for the flutter migrate command
+migration:
+  platforms:
+    - platform: root
+      create_revision: 761747bfc538b5af34aa0d3fac380f1bc331ec49
+      base_revision: 761747bfc538b5af34aa0d3fac380f1bc331ec49
+    - platform: android
+      create_revision: 761747bfc538b5af34aa0d3fac380f1bc331ec49
+      base_revision: 761747bfc538b5af34aa0d3fac380f1bc331ec49
+
+  # User provided section
+
+  # List of Local paths (relative to this file) that should be
+  # ignored by the migrate tool.
+  #
+  # Files that are not part of the templates will be ignored by default.
+  unmanaged_files:
+    - 'lib/main.dart'
+    - 'ios/Runner.xcodeproj/project.pbxproj'
--- a/mobile/plugins/onnx_dart/README.md
+++ b/mobile/plugins/onnx_dart/README.md
@@ -0,0 +1,15 @@
+# onnx_dart
+
+A new Flutter plugin project.
+
+## Getting Started
+
+This project is a starting point for a Flutter
+[plug-in package](https://flutter.dev/developing-packages/),
+a specialized package that includes platform-specific implementation code for
+Android and/or iOS.
+
+For help getting started with Flutter development, view the
+[online documentation](https://flutter.dev/docs), which offers tutorials,
+samples, guidance on mobile development, and a full API reference.
+
--- a/mobile/plugins/onnx_dart/analysis_options.yaml
+++ b/mobile/plugins/onnx_dart/analysis_options.yaml
@@ -0,0 +1 @@
+include: ../../analysis_options.yaml
--- a/mobile/plugins/onnx_dart/android/build.gradle
+++ b/mobile/plugins/onnx_dart/android/build.gradle
@@ -0,0 +1,70 @@
+group = "io.ente.photos.onnx_dart"
+version = "1.0-SNAPSHOT"
+
+buildscript {
+    ext.kotlin_version = "1.7.10"
+    repositories {
+        google()
+        mavenCentral()
+    }
+
+    dependencies {
+        classpath("com.android.tools.build:gradle:7.3.0")
+        classpath("org.jetbrains.kotlin:kotlin-gradle-plugin:$kotlin_version")
+    }
+}
+
+allprojects {
+    repositories {
+        google()
+        mavenCentral()
+    }
+}
+
+apply plugin: "com.android.library"
+apply plugin: "kotlin-android"
+
+android {
+    if (project.android.hasProperty("namespace")) {
+        namespace = "io.ente.photos.onnx_dart"
+    }
+
+    compileSdk = 34
+
+    compileOptions {
+        sourceCompatibility = JavaVersion.VERSION_1_8
+        targetCompatibility = JavaVersion.VERSION_1_8
+    }
+
+    kotlinOptions {
+        jvmTarget = "1.8"
+    }
+
+    sourceSets {
+        main.java.srcDirs += "src/main/kotlin"
+        test.java.srcDirs += "src/test/kotlin"
+    }
+
+    defaultConfig {
+        minSdk = 21
+    }
+
+    dependencies {
+        testImplementation("org.jetbrains.kotlin:kotlin-test")
+        testImplementation("org.mockito:mockito-core:5.0.0")
+
+        implementation 'com.microsoft.onnxruntime:onnxruntime-android:latest.release'
+    }
+
+    testOptions {
+        unitTests.all {
+            useJUnitPlatform()
+
+            testLogging {
+               events "passed", "skipped", "failed", "standardOut", "standardError"
+               outputs.upToDateWhen {false}
+               showStandardStreams = true
+            }
+        }
+    }
+}
--- a/mobile/plugins/onnx_dart/android/local.properties
+++ b/mobile/plugins/onnx_dart/android/local.properties
@@ -0,0 +1,2 @@
+sdk.dir=/Users/ua741/Library/Android/sdk
+flutter.sdk=/Users/ua741/work/flutter
--- a/mobile/plugins/onnx_dart/android/settings.gradle
+++ b/mobile/plugins/onnx_dart/android/settings.gradle
@@ -0,0 +1 @@
+rootProject.name = 'onnx_dart'
--- a/mobile/plugins/onnx_dart/android/src/main/AndroidManifest.xml
+++ b/mobile/plugins/onnx_dart/android/src/main/AndroidManifest.xml
@@ -0,0 +1,3 @@
+<manifest xmlns:android="http://schemas.android.com/apk/res/android"
+  package="io.ente.photos.onnx_dart">
+</manifest>
--- a/mobile/plugins/onnx_dart/android/src/main/kotlin/io/ente/photos/onnx_dart/OnnxDartPlugin.kt
+++ b/mobile/plugins/onnx_dart/android/src/main/kotlin/io/ente/photos/onnx_dart/OnnxDartPlugin.kt
@@ -0,0 +1,253 @@
+package io.ente.photos.onnx_dart
+
+import android.content.Context
+import androidx.annotation.NonNull
+import ai.onnxruntime.*
+import io.flutter.embedding.engine.plugins.FlutterPlugin
+import io.flutter.plugin.common.MethodCall
+import io.flutter.plugin.common.MethodChannel
+import io.flutter.plugin.common.MethodChannel.MethodCallHandler
+import io.flutter.plugin.common.MethodChannel.Result
+import kotlinx.coroutines.*
+import java.nio.FloatBuffer
+import java.util.concurrent.ConcurrentHashMap
+import android.util.Log
+import java.io.File
+import java.nio.ByteBuffer
+import java.nio.ByteOrder
+import java.util.concurrent.ConcurrentLinkedQueue
+
+object LongArrayPool {
+  private val poolMap = ConcurrentHashMap<Int, ConcurrentLinkedQueue<LongArray>>()
+
+  fun get(size: Int): LongArray {
+    val pool = getPool(size)
+    return pool.poll() ?: LongArray(size)
+  }
+
+  fun release(array: LongArray) {
+    val pool = getPool(array.size)
+    pool.offer(array)
+  }
+
+  private fun getPool(size: Int): ConcurrentLinkedQueue<LongArray> {
+    var pool = poolMap[size]
+    if (pool == null) {
+      synchronized(poolMap) {
+        pool = poolMap[size]
+        if (pool == null) {
+          pool = ConcurrentLinkedQueue()
+          poolMap[size] = pool!!
+        }
+      }
+    }
+    return pool!!
+  }
+}
+/** OnnxDartPlugin */
+class OnnxDartPlugin: FlutterPlugin, MethodCallHandler {
+  /// The MethodChannel that will the communication between Flutter and native Android
+  ///
+  /// This local reference serves to register the plugin with the Flutter Engine and unregister it
+  /// when the Flutter Engine is detached from the Activity
+  private lateinit var channel : MethodChannel
+
+  private var faceOrtEnv: OrtEnvironment = OrtEnvironment.getEnvironment()
+  private val scope = CoroutineScope(Dispatchers.IO + SupervisorJob())
+  private val sessionMap = ConcurrentHashMap<ModelType, ModelState>()
+  private lateinit var context: Context
+
+  companion object {
+    const val DEFAULT_SESSION_COUNT = 1
+    const val K_INPUT_WIDTH = 640
+    const val K_INPUT_HEIGHT = 640
+    const val K_NUM_CHANNELS = 3
+  }
+
+  enum class ModelType {
+    CLIP_TEXT, CLIP_VISUAL, YOLO_FACE, MOBILENET_FACE
+  }
+
+  data class ModelState(
+    var isInitialized: Boolean = false,
+    val sessionAddresses: ConcurrentHashMap<Int, OrtSession> = ConcurrentHashMap(),
+    // number of sessions that should have been created for given model
+    var sessionsCount: Int = DEFAULT_SESSION_COUNT
+  )
+
+  override fun onAttachedToEngine(flutterPluginBinding: FlutterPlugin.FlutterPluginBinding) {
+    channel = MethodChannel(flutterPluginBinding.binaryMessenger, "onnx_dart")
+    channel.setMethodCallHandler(this)
+    context = flutterPluginBinding.applicationContext
+  }
+
+  override fun onDetachedFromEngine(binding: FlutterPlugin.FlutterPluginBinding) {
+    channel.setMethodCallHandler(null)
+  }
+
+
+  override fun onMethodCall(call: MethodCall, result: Result) {
+    if (call.method == "getPlatformVersion") {
+      result.success("Android ${android.os.Build.VERSION.RELEASE}")
+    }   else if  (call.method ==  "init") {
+      val modelType = call.argument<String>("modelType") ?: run {
+        result.error("INVALID_ARGUMENT", "Model type is missing", null)
+        return
+      }
+      val modelPath = call.argument<String>("modelPath") ?: run {
+        result.error("INVALID_ARGUMENT", "Model path is missing", null)
+        return
+      }
+      val sessionsCount = call.argument<Int>("sessionsCount") ?: DEFAULT_SESSION_COUNT
+      init(ModelType.valueOf(modelType), modelPath, sessionsCount, result)
+    }
+     else if  (call.method ==  "release" ) {
+      val modelType = call.argument<String>("modelType") ?: run {
+        result.error("INVALID_ARGUMENT", "Model type is missing", null)
+        return
+      }
+      release(ModelType.valueOf(modelType), result)
+    }
+    else if  (call.method ==  "predict" ) {
+      val sessionAddress = call.argument<Int>("sessionAddress")
+      val modelType = call.argument<String>("modelType") ?: run {
+        result.error("INVALID_ARGUMENT", "Model type is missing", null)
+        return
+      }
+      val inputDataArray = call.argument<FloatArray>("inputData")
+
+      if (sessionAddress == null || inputDataArray == null) {
+        result.error("INVALID_ARGUMENT", "Session address or input data is missing", null)
+        return
+      }
+//      val inputDataArray = inputData.map { it.toFloat() }.toFloatArray()
+      predict(ModelType.valueOf(modelType), sessionAddress, inputDataArray, result)
+    } else {
+      result.notImplemented()
+    }
+  }
+
+
+  private fun readModelFile(modelPath: String): ByteArray {
+    return File(modelPath).readBytes()
+  }
+
+  private fun init(modelType: ModelType, modelPath: String, sessionsCount: Int, result: Result) {
+    Log.d("OnnxFlutterPlugin", " v: $modelType, path: $modelPath, sessionsCount: $sessionsCount")
+    scope.launch {
+      val modelState: ModelState
+      if (sessionMap.containsKey(modelType)) {
+        modelState = sessionMap[modelType]!!
+      } else {
+        modelState = ModelState()
+        sessionMap[modelType] = modelState
+      }
+      if (!modelState.isInitialized) {
+        for (i in 0 until sessionsCount) {
+          val session = createSession(faceOrtEnv, modelPath)
+          if (session != null) {
+            modelState.sessionAddresses[i] = session
+          }
+        }
+        modelState.isInitialized = true
+        modelState.sessionsCount = sessionsCount
+        withContext(Dispatchers.Main) {
+          Log.d("OnnxFlutterPlugin", "Model initialized: $modelType")
+          result.success(true)
+        }
+      } else {
+        withContext(Dispatchers.Main) {
+          result.success(false)
+        }
+      }
+    }
+  }
+
+  private fun release(modelType: ModelType, result: Result) {
+    scope.launch {
+      val modelState = sessionMap[modelType]
+      modelState?.let {
+        it.sessionAddresses.forEach { entry: Map.Entry<Int, OrtSession> ->
+          entry.value.close()
+        }
+        it.sessionAddresses.clear()
+        it.isInitialized = false
+      }
+      withContext(Dispatchers.Main) {
+        result.success(true)
+      }
+    }
+  }
+
+  private fun predict(modelType: ModelType, sessionAddress: Int, inputData: FloatArray, result: Result) {
+    scope.launch {
+      val modelState = sessionMap[modelType]
+      val session = modelState?.sessionAddresses?.get(sessionAddress)
+      if (session == null) {
+        withContext(Dispatchers.Main) {
+          result.error("SESSION_NOT_FOUND", "Session not found for address: $sessionAddress", null)
+        }
+        return@launch
+      }
+
+      try {
+        val env = OrtEnvironment.getEnvironment()
+        val inputTensorShape = LongArrayPool.get(4).apply {
+          this[0] = 1
+          this[1] = K_NUM_CHANNELS.toLong()
+          this[2] = K_INPUT_HEIGHT.toLong()
+          this[3] = K_INPUT_WIDTH.toLong()
+        }
+        val inputTensor = OnnxTensor.createTensor(env, FloatBuffer.wrap(inputData), inputTensorShape)
+        val inputs = mapOf("input" to inputTensor)
+        val outputs = session.run(inputs)
+        Log.d("OnnxFlutterPlugin", "Output shape: ${outputs.size()}")
+
+
+        inputTensor.close()
+        val totalSize = 1 * 25200 * 16
+        val flatArray = FloatArray(totalSize) { index -> index + 1.0f }
+
+        val outputTensor2 = Array(1) { outerIndex ->
+          Array(25200) { innerIndex ->
+            val startIndex = (outerIndex * 25200 + innerIndex) * 16
+            flatArray.sliceArray(startIndex until startIndex + 16)
+          }
+        }
+        val outputTensor = outputs[0].value as Array<Array<FloatArray>>
+        Log.d("OnnxFlutterPlugin", "Output2 shape: ${outputTensor.size}")
+
+        outputs.close()
+        LongArrayPool.release(inputTensorShape)
+        // Send the result back to the Dart layer
+        val flatList = outputTensor.flatten().flatMap { it.toList() }
+        val flatList2 = outputTensor2.flatten().flatMap { it.toList() }
+
+
+        withContext(Dispatchers.Main) {
+          result.success(flatList)
+//          result.success(flatList2)
+
+        }
+      } catch (e: OrtException) {
+        withContext(Dispatchers.Main) {
+          result.error("PREDICTION_ERROR", "Error during prediction: ${e.message}", null)
+        }
+      }
+    }
+  }
+
+  private fun createSession(env: OrtEnvironment, modalPath: String): OrtSession? {
+    return env.createSession(modalPath, OrtSession.SessionOptions())
+  }
+
+  private fun releaseAllSessions() {
+    sessionMap.forEach { (_, modelState) ->
+      modelState.sessionAddresses.forEach { entry ->
+        entry.value.close()
+      }
+      modelState.sessionAddresses.clear()
+    }
+    sessionMap.clear()
+  }
+}
--- a/mobile/plugins/onnx_dart/lib/onnx_dart.dart
+++ b/mobile/plugins/onnx_dart/lib/onnx_dart.dart
@@ -0,0 +1,27 @@
+import 'dart:typed_data';
+
+import 'package:onnx_dart/onnx_dart_platform_interface.dart';
+
+class OnnxDart {
+  Future<String?> getPlatformVersion() {
+    return OnnxDartPlatform.instance.getPlatformVersion();
+  }
+
+  Future<bool?> init(
+    String modelType,
+    String modelPath, {
+    int sessionsCount = 1,
+  }) {
+    return OnnxDartPlatform.instance
+        .init(modelType, modelPath, sessionsCount: sessionsCount);
+  }
+
+  Future<dynamic?> predict(
+    Float32List inputData,
+    String modelType, {
+    int sessionAddress = 0,
+  }) async {
+    return OnnxDartPlatform.instance
+        .predict(inputData, modelType, sessionAddress: sessionAddress);
+  }
+}
--- a/mobile/plugins/onnx_dart/lib/onnx_dart_method_channel.dart
+++ b/mobile/plugins/onnx_dart/lib/onnx_dart_method_channel.dart
@@ -0,0 +1,73 @@
+import 'package:flutter/foundation.dart';
+import 'package:flutter/services.dart';
+
+import 'onnx_dart_platform_interface.dart';
+
+/// An implementation of [OnnxDartPlatform] that uses method channels.
+class MethodChannelOnnxDart extends OnnxDartPlatform {
+  /// The method channel used to interact with the native platform.
+  @visibleForTesting
+  final methodChannel = const MethodChannel('onnx_dart');
+
+  @override
+  Future<String?> getPlatformVersion() async {
+    final version =
+        await methodChannel.invokeMethod<String>('getPlatformVersion');
+    return version;
+  }
+
+  @override
+  Future<bool?> init(
+    String modelType,
+    String modelPath, {
+    int sessionsCount = 1,
+  }) async {
+    final result = await methodChannel.invokeMethod<bool>('init', {
+      'modelType': modelType,
+      'modelPath': modelPath,
+      'sessionsCount': sessionsCount,
+    });
+    return result;
+  }
+
+  @override
+  Future<bool?> release(String modelType) async {
+    final bool? result = await methodChannel
+        .invokeMethod<bool>('release', {'modelType': modelType});
+    return result;
+  }
+
+  // @override
+  // Future<List<double>?> predict(
+  //   List<double> inputData,
+  //   String modelType, {
+  //   int sessionAddress = 0,
+  // }) async {
+  //   final List<dynamic>? result =
+  //       await methodChannel.invokeMethod<List<double>?>(
+  //     'predict',
+  //     {
+  //       'sessionAddress': sessionAddress,
+  //       'inputData': inputData,
+  //       'modelType': modelType,
+  //     },
+  //   );
+  //   return result!.cast<double>();
+  // }
+
+  @override
+  Future<dynamic?> predict(
+    Float32List inputData,
+    String modelType, {
+    int sessionAddress = 0,
+  }) {
+    return methodChannel.invokeMethod<dynamic?>(
+      'predict',
+      {
+        'sessionAddress': sessionAddress,
+        'inputData': inputData,
+        'modelType': modelType,
+      },
+    );
+  }
+}
--- a/mobile/plugins/onnx_dart/lib/onnx_dart_platform_interface.dart
+++ b/mobile/plugins/onnx_dart/lib/onnx_dart_platform_interface.dart
@@ -0,0 +1,51 @@
+import 'dart:typed_data';
+
+import 'package:plugin_platform_interface/plugin_platform_interface.dart';
+
+import 'onnx_dart_method_channel.dart';
+
+abstract class OnnxDartPlatform extends PlatformInterface {
+  /// Constructs a OnnxDartPlatform.
+  OnnxDartPlatform() : super(token: _token);
+
+  static final Object _token = Object();
+
+  static OnnxDartPlatform _instance = MethodChannelOnnxDart();
+
+  /// The default instance of [OnnxDartPlatform] to use.
+  ///
+  /// Defaults to [MethodChannelOnnxDart].
+  static OnnxDartPlatform get instance => _instance;
+
+  /// Platform-specific implementations should set this with their own
+  /// platform-specific class that extends [OnnxDartPlatform] when
+  /// they register themselves.
+  static set instance(OnnxDartPlatform instance) {
+    PlatformInterface.verifyToken(instance, _token);
+    _instance = instance;
+  }
+
+  Future<String?> getPlatformVersion() {
+    throw UnimplementedError('platformVersion() has not been implemented.');
+  }
+
+  Future<bool?> init(
+    String modelType,
+    String modelPath, {
+    int sessionsCount = 1,
+  }) {
+    throw UnimplementedError('init() has not been implemented.');
+  }
+
+  Future<bool?> release(String modelType) {
+    throw UnimplementedError('release() has not been implemented.');
+  }
+
+  Future<dynamic?> predict(
+    Float32List inputData,
+    String modelType, {
+    int sessionAddress = 0,
+  }) {
+    throw UnimplementedError('predict() has not been implemented.');
+  }
+}
--- a/mobile/plugins/onnx_dart/pubspec.lock
+++ b/mobile/plugins/onnx_dart/pubspec.lock
@@ -0,0 +1,213 @@
+# Generated by pub
+# See https://dart.dev/tools/pub/glossary#lockfile
+packages:
+  async:
+    dependency: transitive
+    description:
+      name: async
+      sha256: "947bfcf187f74dbc5e146c9eb9c0f10c9f8b30743e341481c1e2ed3ecc18c20c"
+      url: "https://pub.dev"
+    source: hosted
+    version: "2.11.0"
+  boolean_selector:
+    dependency: transitive
+    description:
+      name: boolean_selector
+      sha256: "6cfb5af12253eaf2b368f07bacc5a80d1301a071c73360d746b7f2e32d762c66"
+      url: "https://pub.dev"
+    source: hosted
+    version: "2.1.1"
+  characters:
+    dependency: transitive
+    description:
+      name: characters
+      sha256: "04a925763edad70e8443c99234dc3328f442e811f1d8fd1a72f1c8ad0f69a605"
+      url: "https://pub.dev"
+    source: hosted
+    version: "1.3.0"
+  clock:
+    dependency: transitive
+    description:
+      name: clock
+      sha256: cb6d7f03e1de671e34607e909a7213e31d7752be4fb66a86d29fe1eb14bfb5cf
+      url: "https://pub.dev"
+    source: hosted
+    version: "1.1.1"
+  collection:
+    dependency: transitive
+    description:
+      name: collection
+      sha256: ee67cb0715911d28db6bf4af1026078bd6f0128b07a5f66fb2ed94ec6783c09a
+      url: "https://pub.dev"
+    source: hosted
+    version: "1.18.0"
+  fake_async:
+    dependency: transitive
+    description:
+      name: fake_async
+      sha256: "511392330127add0b769b75a987850d136345d9227c6b94c96a04cf4a391bf78"
+      url: "https://pub.dev"
+    source: hosted
+    version: "1.3.1"
+  flutter:
+    dependency: "direct main"
+    description: flutter
+    source: sdk
+    version: "0.0.0"
+  flutter_lints:
+    dependency: "direct dev"
+    description:
+      name: flutter_lints
+      sha256: "9e8c3858111da373efc5aa341de011d9bd23e2c5c5e0c62bccf32438e192d7b1"
+      url: "https://pub.dev"
+    source: hosted
+    version: "3.0.2"
+  flutter_test:
+    dependency: "direct dev"
+    description: flutter
+    source: sdk
+    version: "0.0.0"
+  leak_tracker:
+    dependency: transitive
+    description:
+      name: leak_tracker
+      sha256: "7f0df31977cb2c0b88585095d168e689669a2cc9b97c309665e3386f3e9d341a"
+      url: "https://pub.dev"
+    source: hosted
+    version: "10.0.4"
+  leak_tracker_flutter_testing:
+    dependency: transitive
+    description:
+      name: leak_tracker_flutter_testing
+      sha256: "06e98f569d004c1315b991ded39924b21af84cf14cc94791b8aea337d25b57f8"
+      url: "https://pub.dev"
+    source: hosted
+    version: "3.0.3"
+  leak_tracker_testing:
+    dependency: transitive
+    description:
+      name: leak_tracker_testing
+      sha256: "6ba465d5d76e67ddf503e1161d1f4a6bc42306f9d66ca1e8f079a47290fb06d3"
+      url: "https://pub.dev"
+    source: hosted
+    version: "3.0.1"
+  lints:
+    dependency: transitive
+    description:
+      name: lints
+      sha256: cbf8d4b858bb0134ef3ef87841abdf8d63bfc255c266b7bf6b39daa1085c4290
+      url: "https://pub.dev"
+    source: hosted
+    version: "3.0.0"
+  matcher:
+    dependency: transitive
+    description:
+      name: matcher
+      sha256: d2323aa2060500f906aa31a895b4030b6da3ebdcc5619d14ce1aada65cd161cb
+      url: "https://pub.dev"
+    source: hosted
+    version: "0.12.16+1"
+  material_color_utilities:
+    dependency: transitive
+    description:
+      name: material_color_utilities
+      sha256: "0e0a020085b65b6083975e499759762399b4475f766c21668c4ecca34ea74e5a"
+      url: "https://pub.dev"
+    source: hosted
+    version: "0.8.0"
+  meta:
+    dependency: transitive
+    description:
+      name: meta
+      sha256: "7687075e408b093f36e6bbf6c91878cc0d4cd10f409506f7bc996f68220b9136"
+      url: "https://pub.dev"
+    source: hosted
+    version: "1.12.0"
+  path:
+    dependency: transitive
+    description:
+      name: path
+      sha256: "087ce49c3f0dc39180befefc60fdb4acd8f8620e5682fe2476afd0b3688bb4af"
+      url: "https://pub.dev"
+    source: hosted
+    version: "1.9.0"
+  plugin_platform_interface:
+    dependency: "direct main"
+    description:
+      name: plugin_platform_interface
+      sha256: "4820fbfdb9478b1ebae27888254d445073732dae3d6ea81f0b7e06d5dedc3f02"
+      url: "https://pub.dev"
+    source: hosted
+    version: "2.1.8"
+  sky_engine:
+    dependency: transitive
+    description: flutter
+    source: sdk
+    version: "0.0.99"
+  source_span:
+    dependency: transitive
+    description:
+      name: source_span
+      sha256: "53e943d4206a5e30df338fd4c6e7a077e02254531b138a15aec3bd143c1a8b3c"
+      url: "https://pub.dev"
+    source: hosted
+    version: "1.10.0"
+  stack_trace:
+    dependency: transitive
+    description:
+      name: stack_trace
+      sha256: "73713990125a6d93122541237550ee3352a2d84baad52d375a4cad2eb9b7ce0b"
+      url: "https://pub.dev"
+    source: hosted
+    version: "1.11.1"
+  stream_channel:
+    dependency: transitive
+    description:
+      name: stream_channel
+      sha256: ba2aa5d8cc609d96bbb2899c28934f9e1af5cddbd60a827822ea467161eb54e7
+      url: "https://pub.dev"
+    source: hosted
+    version: "2.1.2"
+  string_scanner:
+    dependency: transitive
+    description:
+      name: string_scanner
+      sha256: "556692adab6cfa87322a115640c11f13cb77b3f076ddcc5d6ae3c20242bedcde"
+      url: "https://pub.dev"
+    source: hosted
+    version: "1.2.0"
+  term_glyph:
+    dependency: transitive
+    description:
+      name: term_glyph
+      sha256: a29248a84fbb7c79282b40b8c72a1209db169a2e0542bce341da992fe1bc7e84
+      url: "https://pub.dev"
+    source: hosted
+    version: "1.2.1"
+  test_api:
+    dependency: transitive
+    description:
+      name: test_api
+      sha256: "9955ae474176f7ac8ee4e989dadfb411a58c30415bcfb648fa04b2b8a03afa7f"
+      url: "https://pub.dev"
+    source: hosted
+    version: "0.7.0"
+  vector_math:
+    dependency: transitive
+    description:
+      name: vector_math
+      sha256: "80b3257d1492ce4d091729e3a67a60407d227c27241d6927be0130c98e741803"
+      url: "https://pub.dev"
+    source: hosted
+    version: "2.1.4"
+  vm_service:
+    dependency: transitive
+    description:
+      name: vm_service
+      sha256: "3923c89304b715fb1eb6423f017651664a03bf5f4b29983627c4da791f74a4ec"
+      url: "https://pub.dev"
+    source: hosted
+    version: "14.2.1"
+sdks:
+  dart: ">=3.4.3 <4.0.0"
+  flutter: ">=3.18.0-18.0.pre.54"
--- a/mobile/plugins/onnx_dart/pubspec.yaml
+++ b/mobile/plugins/onnx_dart/pubspec.yaml
@@ -0,0 +1,70 @@
+name: onnx_dart
+description: "A new Flutter plugin project."
+version: 0.0.1
+homepage:
+
+environment:
+  sdk: '>=3.4.3 <4.0.0'
+  flutter: '>=3.3.0'
+
+dependencies:
+  flutter:
+    sdk: flutter
+  plugin_platform_interface: ^2.0.2
+
+dev_dependencies:
+  flutter_test:
+    sdk: flutter
+  flutter_lints: ^3.0.0
+
+# For information on the generic Dart part of this file, see the
+# following page: https://dart.dev/tools/pub/pubspec
+
+# The following section is specific to Flutter packages.
+flutter:
+  # This section identifies this Flutter project as a plugin project.
+  # The 'pluginClass' specifies the class (in Java, Kotlin, Swift, Objective-C, etc.)
+  # which should be registered in the plugin registry. This is required for
+  # using method channels.
+  # The Android 'package' specifies package in which the registered class is.
+  # This is required for using method channels on Android.
+  # The 'ffiPlugin' specifies that native code should be built and bundled.
+  # This is required for using `dart:ffi`.
+  # All these are used by the tooling to maintain consistency when
+  # adding or updating assets for this project.
+  plugin:
+    platforms:
+      android:
+        package: io.ente.photos.onnx_dart
+        pluginClass: OnnxDartPlugin
+
+  # To add assets to your plugin package, add an assets section, like this:
+  # assets:
+  #   - images/a_dot_burr.jpeg
+  #   - images/a_dot_ham.jpeg
+  #
+  # For details regarding assets in packages, see
+  # https://flutter.dev/assets-and-images/#from-packages
+  #
+  # An image asset can refer to one or more resolution-specific "variants", see
+  # https://flutter.dev/assets-and-images/#resolution-aware
+
+  # To add custom fonts to your plugin package, add a fonts section here,
+  # in this "flutter" section. Each entry in this list should have a
+  # "family" key with the font family name, and a "fonts" key with a
+  # list giving the asset and other descriptors for the font. For
+  # example:
+  # fonts:
+  #   - family: Schyler
+  #     fonts:
+  #       - asset: fonts/Schyler-Regular.ttf
+  #       - asset: fonts/Schyler-Italic.ttf
+  #         style: italic
+  #   - family: Trajan Pro
+  #     fonts:
+  #       - asset: fonts/TrajanPro.ttf
+  #       - asset: fonts/TrajanPro_Bold.ttf
+  #         weight: 700
+  #
+  # For details regarding fonts in packages, see
+  # https://flutter.dev/custom-fonts/#from-packages
--- a/mobile/pubspec.lock
+++ b/mobile/pubspec.lock
@@ -1606,6 +1606,13 @@ packages:
      url: "https://pub.dev"
    source: hosted
    version: "2.0.0"
+  onnx_dart:
+    dependency: "direct main"
+    description:
+      path: "plugins/onnx_dart"
+      relative: true
+    source: path
+    version: "0.0.1"
  onnxruntime:
    dependency: "direct main"
    description:
@@ -2775,5 +2782,5 @@ packages:
    source: hosted
    version: "3.1.2"
 sdks:
-  dart: ">=3.4.0 <4.0.0"
+  dart: ">=3.4.3 <4.0.0"
  flutter: ">=3.22.0"
--- a/mobile/pubspec.yaml
+++ b/mobile/pubspec.yaml
@@ -128,6 +128,8 @@ dependencies:
    git:
      url: https://github.com/ente-io/onnxruntime.git
      ref: ente_onnxruntime
+  onnx_dart:
+    path: plugins/onnx_dart
  open_mail_app: ^0.4.5
  package_info_plus: ^4.1.0
  page_transition: ^2.0.2