ImageNet Integration

/**
 * Helper class with methods for returning ImageNet label descriptions and
 * decoding prediction arrays to human-readable format.
 */
class ImageNetLabels {
    /**
     * Creates ImageNetLabels instance and loads label data
     * Downloads class labels from remote JSON if not already cached
     */
    ImageNetLabels();
    
    /**
     * Returns the description of the nth class in the 1000 ImageNet classes
     * @param n Class index (0-999)
     * @return String description of the ImageNet class
     */
    String getLabel(int n);
    
    /**
     * Decodes prediction array to top 5 matches with probabilities
     * Given predictions from trained model, returns formatted string
     * listing the top five matches and their respective probabilities
     * @param predictions INDArray containing model predictions
     * @return Formatted string with top 5 predictions and probabilities
     */
    String decodePredictions(INDArray predictions);
}

// Create ImageNet labels helper
ImageNetLabels imageNetLabels = new ImageNetLabels();

// Get specific class label
String label283 = imageNetLabels.getLabel(283); // "Persian cat"
String label285 = imageNetLabels.getLabel(285); // "Egyptian cat"

// Decode model predictions
VGG16 vgg16 = new VGG16(1000, 42, 1);
Model model = vgg16.initPretrained(PretrainedType.IMAGENET);

// Assuming you have preprocessed image data as INDArray input
INDArray predictions = model.output(imageInput);

// Decode to human-readable format
String topPredictions = imageNetLabels.decodePredictions(predictions);
// Output format:
// Predictions for batch :
//   85.234%, Egyptian cat
//   12.456%, Persian cat  
//   1.789%, tabby cat
//   0.345%, tiger cat
//   0.176%, lynx

import org.deeplearning4j.zoo.model.VGG16;
import org.deeplearning4j.zoo.util.imagenet.ImageNetLabels;
import org.deeplearning4j.zoo.PretrainedType;
import org.nd4j.linalg.api.ndarray.INDArray;

// 1. Load pre-trained ImageNet model
VGG16 vgg16 = new VGG16(1000, 42, 1);
Model model = vgg16.initPretrained(PretrainedType.IMAGENET);

// 2. Create ImageNet labels decoder
ImageNetLabels imageNetLabels = new ImageNetLabels();

// 3. Preprocess your image (not shown - requires image loading and preprocessing)
// INDArray imageInput = preprocessImage(imagePath);

// 4. Get model predictions
INDArray predictions = model.output(imageInput);

// 5. Decode predictions to readable format
String results = imageNetLabels.decodePredictions(predictions);
System.out.println(results);

// 6. Get individual class labels
int topClass = predictions.argMax(1).getInt(0);
String topClassName = imageNetLabels.getLabel(topClass);
System.out.println("Top prediction: " + topClassName);

ImageNetLabels labels = new ImageNetLabels();

// Process multiple images in batch
INDArray batchPredictions = model.output(batchInput); // Shape: [batchSize, 1000]

// Decode entire batch - shows results for each image in batch
String batchResults = labels.decodePredictions(batchPredictions);
System.out.println(batchResults);
// Output format for batch:
// Predictions for batch 0 :
//   85.234%, Egyptian cat
//   ...
// Predictions for batch 1 :
//   67.891%, golden retriever
//   ...

ImageNetLabels labels = new ImageNetLabels();

// Common ImageNet classes examples:
String label0 = labels.getLabel(0);     // "tench"
String label1 = labels.getLabel(1);     // "goldfish"
String label151 = labels.getLabel(151); // "Chihuahua"
String label285 = labels.getLabel(285); // "Egyptian cat"
String label945 = labels.getLabel(945); // "bell pepper"

// ImageNet has 1000 classes (indices 0-999)
for (int i = 0; i < 1000; i++) {
    String className = labels.getLabel(i);
    System.out.println("Class " + i + ": " + className);
}

ImageNetLabels labels = new ImageNetLabels();

try {
    // Get label for valid index
    String validLabel = labels.getLabel(500); // Returns valid class name
    
    // Note: Invalid indices may cause exceptions
    // Always ensure index is within 0-999 range for ImageNet
    
} catch (Exception e) {
    System.err.println("Error accessing ImageNet labels: " + e.getMessage());
}

// Works with any ImageNet pre-trained model from the zoo
ImageNetLabels labels = new ImageNetLabels();

// VGG16 with ImageNet weights
VGG16 vgg16 = new VGG16(1000, 42, 1);
Model vgg16Model = vgg16.initPretrained(PretrainedType.IMAGENET);

// AlexNet (if it had ImageNet weights available)
AlexNet alexNet = new AlexNet(1000, 42, 1);
if (alexNet.pretrainedAvailable(PretrainedType.IMAGENET)) {
    Model alexNetModel = alexNet.initPretrained(PretrainedType.IMAGENET);
    INDArray predictions = alexNetModel.output(input);
    String results = labels.decodePredictions(predictions);
}

// ResNet50 with ImageNet weights  
ResNet50 resNet50 = new ResNet50(1000, 42, 1);
if (resNet50.pretrainedAvailable(PretrainedType.IMAGENET)) {
    Model resNetModel = resNet50.initPretrained(PretrainedType.IMAGENET);
    INDArray predictions = resNetModel.output(input);
    String results = labels.decodePredictions(predictions);
}