make model closeable

bioimage-io · Oct 3, 2024 · d9bce6a · d9bce6a
1 parent 21f9682
commit d9bce6a
Show file tree

Hide file tree

Showing 5 changed files with 148 additions and 143 deletions.
diff --git a/src/main/java/io/bioimage/modelrunner/example/ExampleLoadAndRunAllBmzModels.java b/src/main/java/io/bioimage/modelrunner/example/ExampleLoadAndRunAllBmzModels.java
@@ -128,19 +128,20 @@ public static void main(String[] args) {
 	 */
 	public static <T extends RealType<T> & NativeType<T>, R extends RealType<R> & NativeType<R>>
 	void loadAndRunModel(String modelFolder, ModelDescriptor descriptor) throws Exception {
-		Model model = Model.createBioimageioModel(modelFolder, ENGINES_DIR);
-		model.loadModel();
-		List<Tensor<FloatType>> inputs = createInputs(descriptor);
-		List<Tensor<FloatType>> outputs = createOutputs(descriptor);
-		model.runModel(inputs, outputs);
-		for (Tensor<FloatType> tt : outputs) {
-			if (tt.isEmpty())
-				throw new Exception(descriptor.getName() + ": Output tensor is empty");
+		try (Model model = Model.createBioimageioModel(modelFolder, ENGINES_DIR);) {
+			model.loadModel();
+			List<Tensor<FloatType>> inputs = createInputs(descriptor);
+			List<Tensor<FloatType>> outputs = createOutputs(descriptor);
+			model.runModel(inputs, outputs);
+			for (Tensor<FloatType> tt : outputs) {
+				if (tt.isEmpty())
+					throw new Exception(descriptor.getName() + ": Output tensor is empty");
+			}
+
+			model.close();
+			inputs.stream().forEach(t -> t.close());
+			outputs.stream().forEach(t -> t.close());
 		}
-
-		model.closeModel();
-		inputs.stream().forEach(t -> t.close());
-		outputs.stream().forEach(t -> t.close());
 	}
 
 	/**

diff --git a/src/main/java/io/bioimage/modelrunner/example/ExampleLoadAndRunModel.java b/src/main/java/io/bioimage/modelrunner/example/ExampleLoadAndRunModel.java
@@ -139,7 +139,7 @@ void main(String[] args) throws LoadEngineException, Exception {
 		model.runModel(inputs, outputs);
 		System.out.println(Util.average(Util.asDoubleArray(outputs.get(0).getData())));
 		// The result is stored in the list of tensors "outputs"
-		model.closeModel();
+		model.close();
 		inputs.stream().forEach(t -> t.close());
 		outputs.stream().forEach(t -> t.close());
 		System.out.print("Success!!");

diff --git a/src/main/java/io/bioimage/modelrunner/example/ExampleLoadPytorch1Pytorch2.java b/src/main/java/io/bioimage/modelrunner/example/ExampleLoadPytorch1Pytorch2.java
@@ -123,40 +123,40 @@ void loadAndRunPt2(String modelFolder, String modelSource) throws LoadEngineExce
 		// REGARD THAT the engine folders need to follow a naming convention
 		EngineInfo engineInfo = createEngineInfo(framework, engineVersion, enginesDir, cpu, gpu);
 		// Load the corresponding model, for Pytorch model_source arg is not needed
-		Model model = loadModel(modelFolder, modelSource, engineInfo);
-		// Create an image that will be the backend of the Input Tensor
-		final ImgFactory< FloatType > imgFactory = new ArrayImgFactory<>( new FloatType() );
-		final Img< FloatType > img1 = imgFactory.create( 1, 1, 512, 512 );
-		// Create the input tensor with the nameand axes given by the rdf.yaml file
-		// and add it to the list of input tensors
-		Tensor<FloatType> inpTensor = Tensor.build("input0", "bcyx", img1);
-		List<Tensor<FloatType>> inputs = new ArrayList<Tensor<FloatType>>();
-		inputs.add(inpTensor);
-
-		// Create the output tensors defined in the rdf.yaml file with their corresponding 
-		// name and axes and add them to the output list of tensors.
-		/// Regard that output tensors can be built empty without allocating memory
-		// or allocating memory by creating the tensor with a sample empty image, or by
-		// defining the dimensions and data type
-		/*Tensor<FloatType> outTensor = Tensor.buildBlankTensor("output0", 
-				"bcyx", 
-				new long[] {1, 2, 512, 512}, 
-				new FloatType());*/
-		final Img< FloatType > img2 = imgFactory.create( 1, 2, 512, 512 );
-		Tensor<FloatType> outTensor = Tensor.build("output0", "bcyx", img2);
-		List<Tensor<FloatType>> outputs = new ArrayList<Tensor<FloatType>>();
-		outputs.add(outTensor);
-
-		// Run the model on the input tensors. THe output tensors 
-		// will be rewritten with the result of the execution
-		System.out.println(Util.average(Util.asDoubleArray(outputs.get(0).getData())));
-		model.runModel(inputs, outputs);
-		System.out.println(Util.average(Util.asDoubleArray(outputs.get(0).getData())));
-		// The result is stored in the list of tensors "outputs"
-		model.closeModel();
-		inputs.stream().forEach(t -> t.close());
-		outputs.stream().forEach(t -> t.close());
-		System.out.println("Success running Pytorch 2!!");
+		try (Model model = loadModel(modelFolder, modelSource, engineInfo);) {
+			// Create an image that will be the backend of the Input Tensor
+			final ImgFactory< FloatType > imgFactory = new ArrayImgFactory<>( new FloatType() );
+			final Img< FloatType > img1 = imgFactory.create( 1, 1, 512, 512 );
+			// Create the input tensor with the nameand axes given by the rdf.yaml file
+			// and add it to the list of input tensors
+			Tensor<FloatType> inpTensor = Tensor.build("input0", "bcyx", img1);
+			List<Tensor<FloatType>> inputs = new ArrayList<Tensor<FloatType>>();
+			inputs.add(inpTensor);
+			
+			// Create the output tensors defined in the rdf.yaml file with their corresponding 
+			// name and axes and add them to the output list of tensors.
+			/// Regard that output tensors can be built empty without allocating memory
+			// or allocating memory by creating the tensor with a sample empty image, or by
+			// defining the dimensions and data type
+			/*Tensor<FloatType> outTensor = Tensor.buildBlankTensor("output0", 
+					"bcyx", 
+					new long[] {1, 2, 512, 512}, 
+					new FloatType());*/
+			final Img< FloatType > img2 = imgFactory.create( 1, 2, 512, 512 );
+			Tensor<FloatType> outTensor = Tensor.build("output0", "bcyx", img2);
+			List<Tensor<FloatType>> outputs = new ArrayList<Tensor<FloatType>>();
+			outputs.add(outTensor);
+			
+			// Run the model on the input tensors. THe output tensors 
+			// will be rewritten with the result of the execution
+			System.out.println(Util.average(Util.asDoubleArray(outputs.get(0).getData())));
+			model.runModel(inputs, outputs);
+			System.out.println(Util.average(Util.asDoubleArray(outputs.get(0).getData())));
+			// The result is stored in the list of tensors "outputs"
+			inputs.stream().forEach(t -> t.close());
+			outputs.stream().forEach(t -> t.close());
+			System.out.println("Success running Pytorch 2!!");
+		}
 	}
 
 	/**
@@ -198,40 +198,40 @@ void loadAndRunPt1(String modelFolder, String modelSource) throws LoadEngineExce
 		// REGARD THAT the engine folders need to follow a naming convention
 		EngineInfo engineInfo = createEngineInfo(framework, engineVersion, enginesDir, cpu, gpu);
 		// Load the corresponding model, for Pytorch the arg model_source is not needed
-		Model model = loadModel(modelFolder, modelSource, engineInfo);
-		// Create an image that will be the backend of the Input Tensor
-		final ImgFactory< FloatType > imgFactory = new ArrayImgFactory<>( new FloatType() );
-		final Img< FloatType > img1 = imgFactory.create( 1, 1, 512, 512 );
-		// Create the input tensor with the nameand axes given by the rdf.yaml file
-		// and add it to the list of input tensors
-		Tensor<FloatType> inpTensor = Tensor.build("input0", "bcyx", img1);
-		List<Tensor<FloatType>> inputs = new ArrayList<Tensor<FloatType>>();
-		inputs.add(inpTensor);
-
-		// Create the output tensors defined in the rdf.yaml file with their corresponding 
-		// name and axes and add them to the output list of tensors.
-		/// Regard that output tensors can be built empty without allocating memory
-		// or allocating memory by creating the tensor with a sample empty image, or by
-		// defining the dimensions and data type
-		/*Tensor<FloatType> outTensor = Tensor.buildBlankTensor("output0", 
-				"bcyx", 
-				new long[] {1, 2, 512, 512}, 
-				new FloatType());*/
-		final Img< FloatType > img2 = imgFactory.create( 1, 2, 512, 512 );
-		Tensor<FloatType> outTensor = Tensor.build("output0", "bcyx", img2);
-		List<Tensor<FloatType>> outputs = new ArrayList<Tensor<FloatType>>();
-		outputs.add(outTensor);
-
-		// Run the model on the input tensors. THe output tensors 
-		// will be rewritten with the result of the execution
-		System.out.println(Util.average(Util.asDoubleArray(outputs.get(0).getData())));
-		model.runModel(inputs, outputs);
-		System.out.println(Util.average(Util.asDoubleArray(outputs.get(0).getData())));
-		// The result is stored in the list of tensors "outputs"
-		model.closeModel();
-		inputs.stream().forEach(t -> t.close());
-		outputs.stream().forEach(t -> t.close());
-		System.out.println("Success running Pytorch 1!!");
+		try (Model model = loadModel(modelFolder, modelSource, engineInfo);) {
+			// Create an image that will be the backend of the Input Tensor
+			final ImgFactory< FloatType > imgFactory = new ArrayImgFactory<>( new FloatType() );
+			final Img< FloatType > img1 = imgFactory.create( 1, 1, 512, 512 );
+			// Create the input tensor with the nameand axes given by the rdf.yaml file
+			// and add it to the list of input tensors
+			Tensor<FloatType> inpTensor = Tensor.build("input0", "bcyx", img1);
+			List<Tensor<FloatType>> inputs = new ArrayList<Tensor<FloatType>>();
+			inputs.add(inpTensor);
+			
+			// Create the output tensors defined in the rdf.yaml file with their corresponding 
+			// name and axes and add them to the output list of tensors.
+			/// Regard that output tensors can be built empty without allocating memory
+			// or allocating memory by creating the tensor with a sample empty image, or by
+			// defining the dimensions and data type
+			/*Tensor<FloatType> outTensor = Tensor.buildBlankTensor("output0", 
+					"bcyx", 
+					new long[] {1, 2, 512, 512}, 
+					new FloatType());*/
+			final Img< FloatType > img2 = imgFactory.create( 1, 2, 512, 512 );
+			Tensor<FloatType> outTensor = Tensor.build("output0", "bcyx", img2);
+			List<Tensor<FloatType>> outputs = new ArrayList<Tensor<FloatType>>();
+			outputs.add(outTensor);
+			
+			// Run the model on the input tensors. THe output tensors 
+			// will be rewritten with the result of the execution
+			System.out.println(Util.average(Util.asDoubleArray(outputs.get(0).getData())));
+			model.runModel(inputs, outputs);
+			System.out.println(Util.average(Util.asDoubleArray(outputs.get(0).getData())));
+			// The result is stored in the list of tensors "outputs"
+			inputs.stream().forEach(t -> t.close());
+			outputs.stream().forEach(t -> t.close());
+			System.out.println("Success running Pytorch 1!!");
+		}
 	}
 
 	/**

diff --git a/src/main/java/io/bioimage/modelrunner/example/ExampleLoadTensorflow1Tensorflow2.java b/src/main/java/io/bioimage/modelrunner/example/ExampleLoadTensorflow1Tensorflow2.java
@@ -130,36 +130,37 @@ void loadAndRunTf2() throws LoadEngineException, Exception {
 		// REGARD THAT the engine folders need to follow a naming convention
 		EngineInfo engineInfo = createEngineInfo(framework, engineVersion, enginesDir, cpu, gpu);
 		// Load the corresponding model, for Tensorflow model_source arg is not needed
-		Model model = loadModel(modelFolder, null, engineInfo);
-		// Create an image that will be the backend of the Input Tensor
-		final ImgFactory< FloatType > imgFactory = new ArrayImgFactory<>( new FloatType() );
-		final Img< FloatType > img1 = imgFactory.create( 1, 512, 512, 1 );
-		// Create the input tensor with the nameand axes given by the rdf.yaml file
-		// and add it to the list of input tensors
-		Tensor<FloatType> inpTensor = Tensor.build("input_1", "bxyc", img1);
-		List<Tensor<FloatType>> inputs = new ArrayList<Tensor<FloatType>>();
-		inputs.add(inpTensor);
-
-		// Create the output tensors defined in the rdf.yaml file with their corresponding 
-		// name and axes and add them to the output list of tensors.
-		/// Regard that output tensors can be built empty without allocating memory
-		// or allocating memory by creating the tensor with a sample empty image, or by
-		// defining the dimensions and data type
-		Tensor<FloatType> outTensor0 = Tensor.buildBlankTensor(
-				"conv2d_19", "bxyc", new long[] {1, 512, 512, 3}, new FloatType());
-		List<Tensor<FloatType>> outputs = new ArrayList<Tensor<FloatType>>();
-		outputs.add(outTensor0);
-
-		// Run the model on the input tensors. THe output tensors 
-		// will be rewritten with the result of the execution
-		System.out.println(Util.average(Util.asDoubleArray(outputs.get(0).getData())));
-		model.runModel(inputs, outputs);
-		System.out.println(Util.average(Util.asDoubleArray(outputs.get(0).getData())));
-		// The result is stored in the list of tensors "outputs"
-		model.closeModel();
-		inputs.stream().forEach(t -> t.close());
-		outputs.stream().forEach(t -> t.close());
-		System.out.println("Success running Tensorflow 2!!");
+		try (Model model = loadModel(modelFolder, null, engineInfo)){
+			// Create an image that will be the backend of the Input Tensor
+			final ImgFactory< FloatType > imgFactory = new ArrayImgFactory<>( new FloatType() );
+			final Img< FloatType > img1 = imgFactory.create( 1, 512, 512, 1 );
+			// Create the input tensor with the nameand axes given by the rdf.yaml file
+			// and add it to the list of input tensors
+			Tensor<FloatType> inpTensor = Tensor.build("input_1", "bxyc", img1);
+			List<Tensor<FloatType>> inputs = new ArrayList<Tensor<FloatType>>();
+			inputs.add(inpTensor);
+
+			// Create the output tensors defined in the rdf.yaml file with their corresponding 
+			// name and axes and add them to the output list of tensors.
+			/// Regard that output tensors can be built empty without allocating memory
+			// or allocating memory by creating the tensor with a sample empty image, or by
+			// defining the dimensions and data type
+			Tensor<FloatType> outTensor0 = Tensor.buildBlankTensor(
+					"conv2d_19", "bxyc", new long[] {1, 512, 512, 3}, new FloatType());
+			List<Tensor<FloatType>> outputs = new ArrayList<Tensor<FloatType>>();
+			outputs.add(outTensor0);
+
+			// Run the model on the input tensors. THe output tensors 
+			// will be rewritten with the result of the execution
+			System.out.println(Util.average(Util.asDoubleArray(outputs.get(0).getData())));
+			model.runModel(inputs, outputs);
+			System.out.println(Util.average(Util.asDoubleArray(outputs.get(0).getData())));
+			// The result is stored in the list of tensors "outputs"
+			model.close();
+			inputs.stream().forEach(t -> t.close());
+			outputs.stream().forEach(t -> t.close());
+			System.out.println("Success running Tensorflow 2!!");
+		}
 	}
 
 	/**
@@ -202,36 +203,37 @@ void loadAndRunTf1() throws LoadEngineException, Exception {
 		// REGARD THAT the engine folders need to follow a naming convention
 		EngineInfo engineInfo = createEngineInfo(framework, engineVersion, enginesDir, cpu, gpu);
 		// Load the corresponding model, for Tensorflow the arg model_source is not needed
-		Model model = loadModel(modelFolder, null, engineInfo);
-		// Create an image that will be the backend of the Input Tensor
-		final ImgFactory< FloatType > imgFactory = new ArrayImgFactory<>( new FloatType() );
-		final Img< FloatType > img1 = imgFactory.create( 1, 512, 512, 3 );
-		// Create the input tensor with the nameand axes given by the rdf.yaml file
-		// and add it to the list of input tensors
-		Tensor<FloatType> inpTensor = Tensor.build("input", "byxc", img1);
-		List<Tensor<FloatType>> inputs = new ArrayList<Tensor<FloatType>>();
-		inputs.add(inpTensor);
-
-		// Create the output tensors defined in the rdf.yaml file with their corresponding 
-		// name and axes and add them to the output list of tensors.
-		/// Regard that output tensors can be built empty without allocating memory
-		// or allocating memory by creating the tensor with a sample empty image, or by
-		// defining the dimensions and data type
-		final Img< FloatType > img2 = imgFactory.create( 1, 512, 512, 33 );
-		Tensor<FloatType> outTensor = Tensor.build("output", "byxc", img2);
-		List<Tensor<FloatType>> outputs = new ArrayList<Tensor<FloatType>>();
-		outputs.add(outTensor);
-
-		// Run the model on the input tensors. THe output tensors 
-		// will be rewritten with the result of the execution
-		System.out.println(Util.average(Util.asDoubleArray(outputs.get(0).getData())));
-		model.runModel(inputs, outputs);
-		System.out.println(Util.average(Util.asDoubleArray(outputs.get(0).getData())));
-		// The result is stored in the list of tensors "outputs"
-		model.closeModel();
-		inputs.stream().forEach(t -> t.close());
-		outputs.stream().forEach(t -> t.close());
-		System.out.println("Success running Tensorflow 1!!");
+		try (Model model = loadModel(modelFolder, null, engineInfo);){
+			// Create an image that will be the backend of the Input Tensor
+			final ImgFactory< FloatType > imgFactory = new ArrayImgFactory<>( new FloatType() );
+			final Img< FloatType > img1 = imgFactory.create( 1, 512, 512, 3 );
+			// Create the input tensor with the nameand axes given by the rdf.yaml file
+			// and add it to the list of input tensors
+			Tensor<FloatType> inpTensor = Tensor.build("input", "byxc", img1);
+			List<Tensor<FloatType>> inputs = new ArrayList<Tensor<FloatType>>();
+			inputs.add(inpTensor);
+
+			// Create the output tensors defined in the rdf.yaml file with their corresponding 
+			// name and axes and add them to the output list of tensors.
+			/// Regard that output tensors can be built empty without allocating memory
+			// or allocating memory by creating the tensor with a sample empty image, or by
+			// defining the dimensions and data type
+			final Img< FloatType > img2 = imgFactory.create( 1, 512, 512, 33 );
+			Tensor<FloatType> outTensor = Tensor.build("output", "byxc", img2);
+			List<Tensor<FloatType>> outputs = new ArrayList<Tensor<FloatType>>();
+			outputs.add(outTensor);
+
+			// Run the model on the input tensors. THe output tensors 
+			// will be rewritten with the result of the execution
+			System.out.println(Util.average(Util.asDoubleArray(outputs.get(0).getData())));
+			model.runModel(inputs, outputs);
+			System.out.println(Util.average(Util.asDoubleArray(outputs.get(0).getData())));
+			// The result is stored in the list of tensors "outputs"
+			model.close();
+			inputs.stream().forEach(t -> t.close());
+			outputs.stream().forEach(t -> t.close());
+			System.out.println("Success running Tensorflow 1!!");
+		}
 	}
 
 	/**

diff --git a/src/main/java/io/bioimage/modelrunner/model/Model.java b/src/main/java/io/bioimage/modelrunner/model/Model.java
@@ -22,6 +22,7 @@
  */
 package io.bioimage.modelrunner.model;
 
+import java.io.Closeable;
 import java.io.File;
 import java.io.FileNotFoundException;
 import java.io.IOException;
@@ -68,7 +69,7 @@
  * 
  * @author Carlos Garcia Lopez de Haro
  */
-public class Model
+public class Model implements Closeable
 {
 	/**
 	 * Whether the model is loaded or not
@@ -499,10 +500,11 @@ public void loadModel() throws LoadModelException
 	}
 
 	/**
+	 * @Override
 	 * Close the Deep LEarning model in the ClassLoader where the Deep Learning
 	 * framework has been called and instantiated
 	 */
-	public void closeModel()
+	public void close()
 	{
 		DeepLearningEngineInterface engineInstance = getEngineClassLoader().getEngineInstance();
 		engineClassLoader.setEngineClassLoader();