C Deep Neural Network (cdnn) API

NOTICE

The current interface only supports multi-core CPU computing on different platforms (based on openBLAS interface), and support for graphics computation acceleration on different platforms will be added later. (Nvidia's CUDA is expected to be supported first because it provides a C interface, while Apple Silicon currently only offers a C++ interface.)

This project is currently being developed by me alone out of personal interest, and I am just a recent college graduate. Therefore, there may be many issues with this project. I hope more interested people will join me.

This API is still under development and is not yet ready for production use. Please use it with caution and report any issues or bugs you encounter.

An API Test Usage

How to run the test example? (After cloning the project...)

make clean && make && make run_example

(Notice: You need to install libpng, libjpeg-turbo, stb_image, openblas, hdf5, json-c, dirent(for Windows), build-essential(for Linux))

Project Introduction

This project aims to provide a high-level API for designing, training, and evaluating Deep Neural Network models in C. The API is designed to be user-friendly and easy to integrate into existing C projects. It includes functions for creating models, adding layers, compiling models, training models, evaluating models, and saving/loading models and weights (with HDF5 Format).

Code Example:

#include "CNN.h"
#include "dataset.h"

int main() {
    // Create a new CNN model
    Model* model = create(28, 28, 1, 1, 1, 10);

    // Add layers
    add_convolutional_layer(model, 32, 3, 1, 1, "relu");
    add_max_pooling_layer(model, 2, 2);
    add_fully_connected_layer(model, 128, "relu");
    add_dropout_layer(model, 0.5);
    add_fully_connected_layer(model, 10, "softmax");

    // Compile the model
    ModelConfig config = {"Adam", 0.001f, "categorical_crossentropy", "accuracy"};
    compile(model, config);

    // Load datasets
    Dataset* train_dataset = load_dataset("train_data.json");
    Dataset* test_dataset = load_dataset("test_data.json");

    // Train the model
    train(model, train_dataset, 10);

    // Evaluate the model
    float accuracy = evaluate(model, test_dataset);
    printf("Accuracy: %.2f%%\n", accuracy * 100);

    // Free memory
    free_model(model);
    free_dataset(train_dataset);
    free_dataset(test_dataset);

    return 0;
}

Standard JSON Dataset Format of this API

Model Config

{
    "input_shape": [28, 28, 1],
    "output_shape": [10],
    "optimizer": "Adam",
    "learning_rate": 0.001,
    "loss_function": "categorical_crossentropy",
    "metric_name": "accuracy",
    "layers": [
        {
            "type": "convolutional",
            "params": {
                "num_filters": 32,
                "filter_size": 3,
                "stride": 1,
                "padding": "same",
                "activation": "relu"
            }
        },
        {
            "type": "pooling",
            "params": {
                "pool_size": 2,
                "stride": 2,
                "pool_type": "max"
            }
        },
        {
            "type": "fully_connected",
            "params": {
                "num_neurons": 10,
                "activation": "relu"
            }
        }
    ]
}

Dataset with val folder

{
  "dataset_name": "test_data_and_val",
  "num_images": 2,
  "images": [
    {
      "file_path": "dataset example/test_data_and_val/0/test.jpeg",
      "label": "0"
    }
,    {
      "file_path": "dataset example/test_data_and_val/1/test.png",
      "label": "1"
    }
  ]
,
  "val_dataset": "val",
  "num_val_images": 2,
  "val_images": [
    {
      "file_path": "dataset example/test_data_and_val/val/0/test.jpeg",
      "label": "0"
    }
,    {
      "file_path": "dataset example/test_data_and_val/val/1/test.png",
      "label": "1"
    }
  ]
}

Dataset without val folder

{
  "dataset_name": "test_data_without_val",
  "num_images": 4,
  "images": [
    {
      "file_path": "dataset example/test_data_without_val/0/test.jpeg",
      "label": "0"
    }
,    {
      "file_path": "dataset example/test_data_without_val/0/IMG_0255_smaller.jpeg",
      "label": "0"
    }
,    {
      "file_path": "dataset example/test_data_without_val/1/test.png",
      "label": "1"
    }
,    {
      "file_path": "dataset example/test_data_without_val/1/1528_1307.png",
      "label": "1"
    }
  ]
,
  "validation_size": 0.50
}

External Libraries

• libpng: Library for reading and writing PNG image files. License: PNG Reference Library License

• libjpeg-turbo: Library for reading and writing JPEG image files. License: IJG License

• json-c: Library for parsing JSON data. License: MIT License

• hdf5: Library for reading and writing HDF5 files. License: HDF5 License

• You could use vcpkg to install them on Windows