static float visualizeNetwork(NeuralNetwork& neuralNetwork, const Image& referenceImage,
const std::string& outputPath)
{
// save the downsampled reference
Image reference = referenceImage;
reference.setPath(rename(outputPath));
reference.save();
NeuronVisualizer visualizer(&neuralNetwork);
Image image = referenceImage;
image.setPath(outputPath);
visualizer.visualizeNeuron(image, 0);
lucius::util::log("TestVisualization") << "Reference response: "
<< neuralNetwork.runInputs(referenceImage.convertToStandardizedMatrix(
neuralNetwork.getInputCount(),
neuralNetwork.getInputBlockingFactor(), image.colorComponents())).toString();
lucius::util::log("TestVisualization") << "Visualized response: "
<< neuralNetwork.runInputs(image.convertToStandardizedMatrix(
neuralNetwork.getInputCount(),
neuralNetwork.getInputBlockingFactor(), image.colorComponents())).toString();
image.save();
return 0.0f;
}
static void visualizeNeuron(const NeuralNetwork& network, Image& image, unsigned int outputNeuron)
{
Matrix matrix;
std::string solverClass = util::KnobDatabase::getKnobValue(
"NeuronVisualizer::SolverClass", "Differentiable");
if(solverClass == "Differentiable")
{
matrix = optimizeWithDerivative(&network, image, outputNeuron);
}
else if(solverClass == "NonDifferentiable")
{
matrix = optimizeWithoutDerivative(&network, image, outputNeuron);
}
else if(solverClass == "Analytical")
{
matrix = optimizeAnalytically(&network, image, outputNeuron);
}
else
{
throw std::runtime_error("Invalid neuron visializer solver class " + solverClass);
}
size_t x = 0;
size_t y = 0;
util::getNearestToSquareFactors(x, y, network.getInputBlockingFactor());
updateImage(image, matrix, x, y);
}
static float testNetwork(NeuralNetwork& neuralNetwork, const Image& image,
float noiseMagnitude, size_t iterations, size_t batchSize,
std::default_random_engine& engine)
{
float accuracy = 0.0f;
iterations = std::max(iterations, 1UL);
lucius::util::log("TestVisualization") << "Testing the accuracy of the trained network.\n";
for(size_t i = 0; i != iterations; ++i)
{
lucius::util::log("TestVisualization") << " Iteration " << i << " out of "
<< iterations << "\n";
ImageVector batch = generateBatch(image, noiseMagnitude,
batchSize, engine);
Matrix input = batch.convertToStandardizedMatrix(
neuralNetwork.getInputCount(),
neuralNetwork.getInputBlockingFactor(), image.colorComponents());
Matrix reference = generateReference(batch);
lucius::util::log("TestVisualization") << " Input: " << input.toString();
lucius::util::log("TestVisualization") << " Reference: " << reference.toString();
accuracy += neuralNetwork.computeAccuracy(input, reference);
}
return accuracy * 100.0f / iterations;
}
static void trainNetwork(NeuralNetwork& neuralNetwork, const Image& image,
float noiseMagnitude, size_t iterations, size_t batchSize,
std::default_random_engine& engine)
{
lucius::util::log("TestVisualization") << "Training the network.\n";
for(size_t i = 0; i != iterations; ++i)
{
lucius::util::log("TestVisualization") << " Iteration " << i << " out of "
<< iterations << "\n";
ImageVector batch = generateBatch(image, noiseMagnitude,
batchSize, engine);
Matrix input = batch.convertToStandardizedMatrix(
neuralNetwork.getInputCount(),
neuralNetwork.getInputBlockingFactor(), image.colorComponents());
Matrix reference = generateReference(batch);
lucius::util::log("TestVisualization") << " Input: " << input.toString();
lucius::util::log("TestVisualization") << " Reference: " << reference.toString();
neuralNetwork.train(input, reference);
}
}
