Do not implement inference model in high-level model classes

api/src/main/kotlin/org/jetbrains/kotlinx/dl/api/inference/loaders/ModelHub.kt

@@ -28,7 +28,7 @@ public abstract class ModelHub {
      * @param [loadingMode] Strategy of existing model use-case handling.
      * @return Raw model without weights. Needs in compilation and weights loading before usage.
      */
-    public abstract fun <T : InferenceModel, U : InferenceModel> loadModel(
+    public abstract fun <T : InferenceModel, U> loadModel(
         modelType: ModelType<T, U>,
         loadingMode: LoadingMode = LoadingMode.SKIP_LOADING_IF_EXISTS
     ): T
@@ -40,7 +40,7 @@ public abstract class ModelHub {
      * @param [loadingMode] Strategy of existing model use-case handling.
      * @return Pretrained model.
      */
-    public fun <T : InferenceModel, U : InferenceModel> loadPretrainedModel(
+    public fun <T : InferenceModel, U> loadPretrainedModel(
         modelType: ModelType<T, U>,
         loadingMode: LoadingMode = LoadingMode.SKIP_LOADING_IF_EXISTS
     ): U {
@@ -50,7 +50,7 @@ public abstract class ModelHub {
     /**
      * This operator equivalent to [loadPretrainedModel].
      */
-    public operator fun <T : InferenceModel, U : InferenceModel> get(modelType: ModelType<T, U>): U {
+    public operator fun <T : InferenceModel, U> get(modelType: ModelType<T, U>): U {
         return loadPretrainedModel(modelType = modelType)
     }
 }

api/src/main/kotlin/org/jetbrains/kotlinx/dl/api/inference/loaders/ModelType.kt

@@ -15,7 +15,7 @@ import org.jetbrains.kotlinx.dl.api.preprocessing.Operation
  * @param T the type of the basic model for common functionality.
  * @param U the type of the pre-trained model for usage in Easy API.
  */
-public interface ModelType<T : InferenceModel, U : InferenceModel> {
+public interface ModelType<T : InferenceModel, U> {
     /** Relative path to model for local and S3 buckets storages. */
     public val modelRelativePath: String
 

impl/src/commonMain/kotlin/org/jetbrains/kotlinx/dl/impl/inference/imagerecognition/ImageRecognitionModelBase.kt

@@ -22,7 +22,7 @@ import org.jetbrains.kotlinx.dl.api.summary.ModelWithSummary
 public abstract class ImageRecognitionModelBase<I>(
     protected val internalModel: InferenceModel,
     protected val modelKindDescription: String? = null
-) : InferenceModel by internalModel, ModelWithSummary {
+) : ModelWithSummary, AutoCloseable {
     /**
      * Preprocessing operation specific to this model.
      */
@@ -70,4 +70,6 @@ public abstract class ImageRecognitionModelBase<I>(
             ModelHubModelSummary(EmptySummary(), modelKindDescription)
         }
     }
+
+    override fun close(): Unit = internalModel.close()
 }

impl/src/jvmMain/kotlin/org/jetbrains/kotlinx/dl/impl/inference/imagerecognition/ImageRecognitionModel.kt

@@ -63,20 +63,6 @@ public class ImageRecognitionModel(
         return predictObject(ImageConverter.toBufferedImage(imageFile))
     }
 
-    override fun copy(
-        copiedModelName: String?,
-        saveOptimizerState: Boolean,
-        copyWeights: Boolean
-    ): ImageRecognitionModel {
-        return ImageRecognitionModel(
-            internalModel.copy(copiedModelName, saveOptimizerState, copyWeights),
-            inputColorMode,
-            channelsFirst,
-            preprocessor,
-            modelKindDescription
-        )
-    }
-
     public companion object {
         /**
          * Creates a preprocessing [Operation] which converts given [BufferedImage] to [FloatData].

onnx/src/androidMain/kotlin/org/jetbrains/kotlinx/dl/onnx/inference/ONNXModels.kt

@@ -106,7 +106,7 @@ public object ONNXModels {
     }
 
     /** Pose detection models. */
-    public sealed class PoseDetection<U : InferenceModel>(override val modelRelativePath: String) :
+    public sealed class PoseDetection<U>(override val modelRelativePath: String) :
         OnnxModelType<U> {
         /**
          * This model is a convolutional neural network model that runs on RGB images and predicts human joint locations of a single person.
@@ -164,7 +164,7 @@ public object ONNXModels {
     }
 
     /** Object detection models and preprocessing. */
-    public sealed class ObjectDetection<U : InferenceModel>(override val modelRelativePath: String) :
+    public sealed class ObjectDetection<U>(override val modelRelativePath: String) :
         OnnxModelType<U> {
         /**
          * This model is a real-time neural network for object detection that detects 90 different classes
@@ -291,7 +291,7 @@ public object ONNXModels {
     }
 
     /** Face alignment models */
-    public sealed class FaceAlignment<U : InferenceModel> : OnnxModelType<U> {
+    public sealed class FaceAlignment<U> : OnnxModelType<U> {
         /**
          * This model is a neural network for face alignment that take RGB images of faces as input and produces coordinates of 106 faces landmarks.
          *

onnx/src/androidMain/kotlin/org/jetbrains/kotlinx/dl/onnx/inference/OnnxModelHub.kt

@@ -43,10 +43,10 @@ public class ONNXModelHub(private val context: Context) : ModelHub() {
      * @param [loadingMode] it's ignored
      */
     @Suppress("UNCHECKED_CAST")
-    override fun <T : InferenceModel, U : InferenceModel> loadModel(
+    override fun <T : InferenceModel, U> loadModel(
         modelType: ModelType<T, U>,
         loadingMode: LoadingMode, /* unused */
     ): T {
-        return loadModel(modelType as OnnxModelType<U>) as T
+        return loadModel(modelType as OnnxModelType<*>) as T
     }
 }

onnx/src/androidMain/kotlin/org/jetbrains/kotlinx/dl/onnx/inference/facealignment/FaceDetectionModel.kt

@@ -28,7 +28,7 @@ import org.jetbrains.kotlinx.dl.onnx.inference.doWithRotation
 public class FaceDetectionModel(
     override val internalModel: OnnxInferenceModel,
     modelKindDescription: String? = null
-) : FaceDetectionModelBase<Bitmap>(modelKindDescription), CameraXCompatibleModel, InferenceModel by internalModel {
+) : FaceDetectionModelBase<Bitmap>(modelKindDescription), CameraXCompatibleModel {
     override var targetRotation: Int = 0
     override val preprocessing: Operation<Bitmap, FloatData>
         get() = pipeline<Bitmap>()
@@ -40,12 +40,7 @@ public class FaceDetectionModel(
             .toFloatArray { layout = TensorLayout.NCHW }
             .call(ONNXModels.FaceDetection.defaultPreprocessor)
 
-    override fun copy(copiedModelName: String?, saveOptimizerState: Boolean, copyWeights: Boolean): InferenceModel {
-        return FaceDetectionModel(
-            internalModel.copy(copiedModelName, saveOptimizerState, copyWeights),
-            modelKindDescription
-        )
-    }
+    override fun close(): Unit = internalModel.close()
 }
 
 /**

onnx/src/androidMain/kotlin/org/jetbrains/kotlinx/dl/onnx/inference/facealignment/Fan2D106FaceAlignmentModel.kt

@@ -29,7 +29,7 @@ import org.jetbrains.kotlinx.dl.onnx.inference.doWithRotation
 public class Fan2D106FaceAlignmentModel(
     override val internalModel: OnnxInferenceModel,
     modelKindDescription: String? = null
-) : FaceAlignmentModelBase<Bitmap>(modelKindDescription), CameraXCompatibleModel, InferenceModel by internalModel {
+) : FaceAlignmentModelBase<Bitmap>(modelKindDescription), CameraXCompatibleModel {
     override val outputName: String = "fc1"
     override var targetRotation: Int = 0
 
@@ -41,12 +41,7 @@ public class Fan2D106FaceAlignmentModel(
         .rotate { degrees = targetRotation.toFloat() }
         .toFloatArray { layout = TensorLayout.NCHW }
 
-    override fun copy(copiedModelName: String?, saveOptimizerState: Boolean, copyWeights: Boolean): InferenceModel {
-        return Fan2D106FaceAlignmentModel(
-            internalModel.copy(copiedModelName, saveOptimizerState, copyWeights),
-            modelKindDescription
-        )
-    }
+    override fun close(): Unit = internalModel.close()
 }
 
 /**

onnx/src/androidMain/kotlin/org/jetbrains/kotlinx/dl/onnx/inference/objectdetection/SSDLikeModel.kt

@@ -36,7 +36,7 @@ import org.jetbrains.kotlinx.dl.onnx.inference.doWithRotation
 public class SSDLikeModel(
     override val internalModel: OnnxInferenceModel, metadata: SSDLikeModelMetadata,
     modelKindDescription: String? = null
-) : SSDLikeModelBase<Bitmap>(metadata, modelKindDescription), CameraXCompatibleModel, InferenceModel by internalModel {
+) : SSDLikeModelBase<Bitmap>(metadata, modelKindDescription), CameraXCompatibleModel {
 
     override val classLabels: Map<Int, String> = Coco.V2017.labels(zeroIndexed = true)
 

onnx/src/androidMain/kotlin/org/jetbrains/kotlinx/dl/onnx/inference/posedetection/SinglePoseDetectionModel.kt

@@ -35,8 +35,7 @@ import org.jetbrains.kotlinx.dl.onnx.inference.executionproviders.ExecutionProvi
 public class SinglePoseDetectionModel(
     override val internalModel: OnnxInferenceModel,
     modelKindDescription: String? = null
-) : SinglePoseDetectionModelBase<Bitmap>(modelKindDescription), InferenceModel by internalModel,
-    CameraXCompatibleModel {
+) : SinglePoseDetectionModelBase<Bitmap>(modelKindDescription), CameraXCompatibleModel {
     override val preprocessing: Operation<Bitmap, FloatData>
         get() = pipeline<Bitmap>()
             .resize {

onnx/src/commonMain/kotlin/org/jetbrains/kotlinx/dl/onnx/inference/OnnxModelType.kt

@@ -5,13 +5,12 @@
 
 package org.jetbrains.kotlinx.dl.onnx.inference
 
-import org.jetbrains.kotlinx.dl.api.inference.InferenceModel
 import org.jetbrains.kotlinx.dl.api.inference.loaders.ModelType
 
 /**
  * Base type for [OnnxInferenceModel].
  */
-public interface OnnxModelType<U : InferenceModel> : ModelType<OnnxInferenceModel, U> {
+public interface OnnxModelType<U> : ModelType<OnnxInferenceModel, U> {
     /**
      * Shape of the input accepted by this model, without batch size.
      */

onnx/src/jvmMain/kotlin/org/jetbrains/kotlinx/dl/onnx/inference/ONNXModelHub.kt

@@ -47,11 +47,11 @@ public class ONNXModelHub(public val cacheDirectory: File) : ModelHub() {
      * @return An instance of [OnnxInferenceModel].
      */
     @Suppress("UNCHECKED_CAST")
-    public override fun <T : InferenceModel, U : InferenceModel> loadModel(
+    public override fun <T : InferenceModel, U> loadModel(
         modelType: ModelType<T, U>,
         loadingMode: LoadingMode
     ): T {
-        return loadModel(modelType as OnnxModelType<U>, ExecutionProvider.CPU(), loadingMode = loadingMode) as T
+        return loadModel(modelType as OnnxModelType<*>, ExecutionProvider.CPU(), loadingMode = loadingMode) as T
     }
 
     private fun getONNXModelFile(modelFile: String, loadingMode: LoadingMode): File {

onnx/src/jvmMain/kotlin/org/jetbrains/kotlinx/dl/onnx/inference/ONNXModels.kt

@@ -579,7 +579,7 @@ public object ONNXModels {
     }
 
     /** Object detection models and preprocessing. */
-    public sealed class ObjectDetection<U : InferenceModel>(override val modelRelativePath: String) :
+    public sealed class ObjectDetection<U>(override val modelRelativePath: String) :
         OnnxModelType<U> {
         /**
          * This model is a real-time neural network for object detection that detects 80 different classes
@@ -895,7 +895,7 @@ public object ONNXModels {
     }
 
     /** Face alignment models and preprocessing. */
-    public sealed class FaceAlignment<U : InferenceModel>(override val modelRelativePath: String) :
+    public sealed class FaceAlignment<U>(override val modelRelativePath: String) :
         OnnxModelType<U> {
         /**
          * This model is a neural network for face alignment that take RGB images of faces as input and produces coordinates of 106 faces landmarks.
@@ -916,7 +916,7 @@ public object ONNXModels {
     }
 
     /** Pose detection models. */
-    public sealed class PoseDetection<U : InferenceModel>(override val modelRelativePath: String) :
+    public sealed class PoseDetection<U>(override val modelRelativePath: String) :
         OnnxModelType<U> {
         /**
          * This model is a convolutional neural network model that runs on RGB images and predicts human joint locations of a single person.

onnx/src/jvmMain/kotlin/org/jetbrains/kotlinx/dl/onnx/inference/facealignment/FaceDetectionModel.kt

@@ -27,7 +27,7 @@ import java.awt.image.BufferedImage
 public class FaceDetectionModel(
     override val internalModel: OnnxInferenceModel,
     modelKindDescription: String? = null
-) : FaceDetectionModelBase<BufferedImage>(modelKindDescription), InferenceModel by internalModel {
+) : FaceDetectionModelBase<BufferedImage>(modelKindDescription) {
     override val preprocessing: Operation<BufferedImage, FloatData>
         get() = pipeline<BufferedImage>()
             .resize {
@@ -38,10 +38,5 @@ public class FaceDetectionModel(
             .toFloatArray { }
             .call(ONNXModels.FaceDetection.defaultPreprocessor)
 
-    override fun copy(copiedModelName: String?, saveOptimizerState: Boolean, copyWeights: Boolean): InferenceModel {
-        return FaceDetectionModel(
-            internalModel.copy(copiedModelName, saveOptimizerState, copyWeights),
-            modelKindDescription
-        )
-    }
+    override fun close(): Unit = internalModel.close()
 }