1. Load and initialize the training data using ImageRecordReader:

ImageRecordReader imageRecordReader = new ImageRecordReader(imageHeight,imageWidth,channels,parentPathLabelGenerator);
 imageRecordReader.initialize(trainData,null);

2. Create a dataset iterator using RecordReaderDataSetIterator:

DataSetIterator dataSetIterator = new RecordReaderDataSetIterator(imageRecordReader,batchSize,1,numLabels);

3. Add the normalizer to the dataset iterator:

DataNormalization scaler = new ImagePreProcessingScaler(0,1);
 scaler.fit(dataSetIterator);
 dataSetIterator.setPreProcessor(scaler);
 

4. Train the model by calling fit():

MultiLayerConfiguration config = builder.build();
 MultiLayerNetwork model = new MultiLayerNetwork(config);
 model.init();
 model.setListeners(new ScoreIterationListener(100));
 model.fit(dataSetIterator,epochs);

5. Train the model again with image transformations:

imageRecordReader.initialize(trainData,transform);
 dataSetIterator = new RecordReaderDataSetIterator(imageRecordReader,batchSize,1,numLabels);
 scaler.fit(dataSetIterator);
 dataSetIterator.setPreProcessor(scaler);
 model.fit(dataSetIterator,epochs);

6. Evaluate the model and observe the results:

Evaluation evaluation = model.evaluate(dataSetIterator);
 System.out.println(evaluation.stats()); 

The evaluation metrics will appear as follows:

7. Add support for the GPU-accelerated environment by adding the following dependencies:

<dependency>
  <groupId>org.nd4j</groupId>
  <artifactId>nd4j-cuda-9.1-platform</artifactId>
  <version>1.0.0-beta3</version>
 </dependency>
 
 <dependency>
  <groupId>org.deeplearning4j</groupId>
  <artifactId>deeplearning4j-cuda-9.1</artifactId>
  <version>1.0.0-beta3</version>
 </dependency>