These examples help you to understand mlboard::SummaryWriter::Image() API in depth.
- Log a single image
- Log multiple images
- Log multiple images stored in arma::mat
- Log images stored at a location
A single Image could be logged using the following API.
template<typename Filewriter>
void SummaryWriter<Filewriter>::Image(const std::string& tag,
int step,
const std::string& encodedImage,
int height,
int width,
int channel,
mlboard::Filewriter& fw,
const std::string& displayName,
const std::string& description)The API accepts tag, step, encodedImage (Image data in encoded format), the metadata about image (height and width) along with mlboard::Filewriter object,Name and description.
Following is a snippet that would log a single image for 1 step in temp directory.
#include <mlboard/mlboard.hpp>
#include <iostream>
#include <chrono>
#include <ctime>
#include <future>
int main()
{
std::chrono::time_point<std::chrono::system_clock> start, end;
start = std::chrono::system_clock::now();
mlboard::FileWriter f1("temp");
std::ifstream fin("./assets/single_image.jpg", std::ios::binary);
std::ostringstream ss;
ss << fin.rdbuf();
std::string image(ss.str());
fin.close();
mlboard::SummaryWriter<mlboard::FileWriter>::Image(
"Test Image", 1, image, 512, 512, 3, f1, "Sample Image",
"This is a Sample image logged using mlboard.");
f1.Close();
end = std::chrono::system_clock::now();
std::chrono::duration<double> elapsed_seconds = end - start;
std::time_t end_time = std::chrono::system_clock::to_time_t(end);
std::cout << "finished computation at " << std::ctime(&end_time)
<< "elapsed time: " << elapsed_seconds.count() << "s\n";
}The output would be similar to:
To log multiple image at an instance you can use the following API:
void SummaryWriter<Filewriter>::Image(const std::string& tag,
int step,
const std::vector<std::string>& encodedImages,
int height,
int width,
mlboard::Filewriter& fw,
const std::string& displayName,
const std::string& description)The API accepts tag, step, encodedImage (A vector of Image data in encoded format), the metadata about images (height and width) along with mlboard::Filewriter object, Name and description.
Following is a snippet that would log a multiple image for 1 step in temp directory.
#include <mlboard/mlboard.hpp>
#include <iostream>
#include <chrono>
#include <ctime>
#include <future>
int main()
{
std::chrono::time_point<std::chrono::system_clock> start, end;
start = std::chrono::system_clock::now();
mlboard::FileWriter f1("temp");
std::ifstream fin("./assets/multiple_image.jpg", std::ios::binary);
std::ostringstream ss;
std::vector<std::string>encodedImages;
ss << fin.rdbuf();
encodedImages.push_back(ss.str());
fin.close();
fin.open("./assets/single_image.jpg", std::ios::binary);
ss << fin.rdbuf();
encodedImages.push_back(ss.str());
ss.str("");
fin.close();
mlboard::SummaryWriter<mlboard::FileWriter>::Image(
"Test Image", 1, encodedImages, 512, 512, f1, "Sample Image",
"This is a Sample image logged using mlboard.");
f1.Close();
end = std::chrono::system_clock::now();
std::chrono::duration<double> elapsed_seconds = end - start;
std::time_t end_time = std::chrono::system_clock::to_time_t(end);
std::cout << "finished computation at " << std::ctime(&end_time)
<< "elapsed time: " << elapsed_seconds.count() << "s\n";
}The output would be similar to:
Since we would be logging mlpack's metrics there are chances when we would have to visualize images stored in arma::mat and thus you can use the following API to directly log Armadillo matrices:
void SummaryWriter<Filewriter>::Image(const std::string& tag,
int step,
arma::Mat<eT>& matrix,
mlpack::data::ImageInfo& info,
mlboard::Filewriter& fw,
const std::string& displayName,
const std::string& description)The API accepts tag, step, encodedImage (An arma::mat which has image data), mlpack::data::ImageInfo object, mlboard::Filewriter object, name and description.
Following is a snippet that would log a multiple image which is stored in arma::mat for 1 step in temp directory.
#include <mlboard/mlboard.hpp>
#include <iostream>
#include <chrono>
#include <ctime>
#include <future>
int main()
{
std::chrono::time_point<std::chrono::system_clock> start, end;
start = std::chrono::system_clock::now();
mlboard::FileWriter f1("temp");
// Following line are just there to come up with a arma::mat of images.
// If you already have a matrix with images you can avoid the following lines.
arma::Mat<unsigned char> matrix;
mlpack::data::ImageInfo info;
std::vector<std::string> files = {"./assets/single_image.jpg",
"./assets/single_image.jpg"};
// Creating the matrix which has image.
mlpack::data::Load(files, matrix, info, false);
// Now we can log the matrix.
mlboard::SummaryWriter<mlboard::FileWriter>::Image(
"Multiple Image", 1, matrix, info, f1, "Sample Multiple Image",
"This is a Sample image logged using mlboard.");
f1.Close();
end = std::chrono::system_clock::now();
std::chrono::duration<double> elapsed_seconds = end - start;
std::time_t end_time = std::chrono::system_clock::to_time_t(end);
std::cout << "finished computation at " << std::ctime(&end_time)
<< "elapsed time: " << elapsed_seconds.count() << "s\n";
}The output would be similar to Log multiple image
Note: This API takes quite a time to log the image since it first has to process the image stored in arma::mat. The way it process is, it saves the image using mlpack::data::Load() API for image and then reads it back in encoded format. It is something similar to what mxnet logging tool does here. Hence it is recommended to log image summary using a arma::mat in a separate thread to avoid an increase in training time. Also, make sure that all the images are of the same height and width since mlpack::load() function only supports loading images of the same sizes.
There are many occasion when a user wants to log images which are there in a current folder and you can use the utility function provided under util namespace along with any of the above api.
The utility allows you to convert an image given its path into Encodedformat which can be logged to a file. The API is:
void EncodeImage(std::vector<std::string>& filePaths,
std::vector<std::string>>& encodedImages)The function accepts a vector of string which has filePath of the image to be converted and also accepts an empty vector where all the encoded images could be saved.
An example using the above utility could be:
#include <mlboard/mlboard.hpp>
#include <iostream>
#include <chrono>
#include <ctime>
#include <future>
int main()
{
std::chrono::time_point<std::chrono::system_clock> start, end;
start = std::chrono::system_clock::now();
mlboard::FileWriter f1("temp");
// Using the utitlity function encodeImages stored at current location
std::vector<std::string>encodedImages;
std::vector<std::string>filePaths = {"./assets/single_image.jpg",
"./assets/multiple_image.jpg"};
mlboard::util::EncodeImage(filePaths, encodedImages);
mlboard::SummaryWriter<mlboard::FileWriter>::Image(
"Test Image", 1, encodedImages, 512, 512, f1, "Sample Image",
"This is a Sample image logged using mlboard.");
f1.Close();
end = std::chrono::system_clock::now();
std::chrono::duration<double> elapsed_seconds = end - start;
std::time_t end_time = std::chrono::system_clock::to_time_t(end);
std::cout << "finished computation at " << std::ctime(&end_time)
<< "elapsed time: " << elapsed_seconds.count() << "s\n";
}