#include <time.h>
#include <nntrainer.h>
+#if defined(__TIZEN__)
+#include <nnstreamer.h>
+#endif
/** Number of dimensions for the input data */
#define MAX_DIM 4
/** Vectors containing the training data */
std::vector<std::vector<float>> inputVector, labelVector;
-unsigned int iteration = 0;
/**
- * @brief step function
- * @param[in] x value to be distinguished
- * @retval 0.0 or 1.0
+ * @brief Private data for Tensorflow lite object
*/
-// float stepFunction(float x) {
-// if (x > 0.9) {
-// return 1.0;
-// }
-//
-// if (x < 0.1) {
-// return 0.0;
-// }
-//
-// return x;
-// }
-
struct TFLiteData {
tflite::ops::builtin::BuiltinOpResolver resolver;
std::unique_ptr<tflite::Interpreter> interpreter;
int inputDimReq[MAX_DIM];
};
+/**
+ * @brief Load the tensorflow lite model and its metadata
+ */
void setupTensorflowLiteModel(const std::string &data_path,
TFLiteData &tflite_data) {
int input_size;
/**
* @brief Extract the features from pretrained model
- * @param[in] data_path data path
+ * @param[in] tflite_data private data for tflite model
* @param[out] input_data output of tflite model (input for the nntrainer model)
* @param[out] label_data one hot label data
*/
* Data generator callback
*/
int getBatch_train(float **input, float **label, bool *last, void *user_data) {
+ static unsigned int iteration = 0;
if (iteration >= EPOCH_SIZE) {
*last = true;
iteration = 0;
}
/**
- * @brief create NN
- * Get Feature from tflite & run foword & back propatation
- * @param[in] arg 1 : configuration file path
- * @param[in] arg 2 : resource path
+ * @brief Train the model with the given config file path
+ * @param[in] config Model config file path
*/
-int main(int argc, char *argv[]) {
+int trainModel(const char *config) {
int status = ML_ERROR_NONE;
- if (argc < 3) {
- std::cout << "./TransferLearning Config.ini resources\n";
- exit(0);
- }
-
- const std::vector<std::string> args(argv + 1, argv + argc);
- std::string config = args[0];
-
- /** location of resources ( ../../res/ ) */
- std::string data_path = args[1];
-
- srand(time(NULL));
-
- TFLiteData tflite_data;
- setupTensorflowLiteModel(data_path, tflite_data);
-
- /** Extract features from the already trained model */
- extractFeatures(tflite_data, inputVector, labelVector);
/** Neural Network Create & Initialization */
ml_train_model_h handle = NULL;
ml_train_dataset_h dataset = NULL;
- status = ml_train_model_construct_with_conf(config.c_str(), &handle);
+ status = ml_train_model_construct_with_conf(config, &handle);
if (status != ML_ERROR_NONE) {
std::cerr << "Failed to construct the model" << std::endl;
return status;
}
/** destroy the model */
- ml_train_model_destroy(handle);
+ status = ml_train_model_destroy(handle);
+ return status;
+}
- /**
- * @brief test
- */
- // for (int i = 0; i < TOTAL_TEST_SIZE; i++) {
- // std::string path = data_path;
- // path += "testset";
- // printf("\n[%s]\n", path.c_str());
- // std::string img = path + "/";
- // img += "test" + std::to_string(i + 1) + ".bmp";
- // printf("%s\n", img.c_str());
-
- // std::vector<float> featureVector, resultVector;
- // featureVector.resize(128);
- // getFeature(img, featureVector);
- // nntrainer::Tensor X;
- // try {
- // X = nntrainer::Tensor({featureVector});
- // NN.forwarding(MAKE_SHARED_TENSOR(X))->apply(stepFunction);
- // } catch (...) {
- // std::cerr << "Error during forwaring the model" << std::endl;
- // NN.finalize();
- // return -1;
- // }
- // }
+/**
+ * @brief Test the model with the given config file path
+ * @param[in] data_path Path of the test data
+ * @param[in] tflite_data TFLite loaded configuration
+ * @param[in] config Model config file path
+ */
+int testModel(std::string data_path, const TFLiteData &tflite_data,
+ const char *model) {
+#if defined(__TIZEN__)
+ int status = ML_ERROR_NONE;
+ ml_pipeline_h pipe;
+ ml_pipeline_src_h src;
+ ml_tensors_info_h in_info;
+ ml_tensors_data_h in_data;
+ void *raw_data;
+ size_t data_size;
+ ml_tensor_dimension in_dim = {1, 1, 1, INPUT_SIZE};
+
+ char pipeline[1024];
+ snprintf(
+ pipeline, sizeof(pipeline),
+ "appsrc name=srcx | "
+ "other/"
+ "tensor,dimension=(string)1:1:1:%d,type=(string)float,framerate=(fraction)"
+ "0/1 | tensor_filter framework=nntrainer model=%s | tensor_sink",
+ INPUT_SIZE, model);
+
+ status = ml_pipeline_construct(pipeline, NULL, NULL, &pipe);
+ if (status != ML_ERROR_NONE)
+ goto fail_exit;
+
+ status = ml_pipeline_src_get_handle(pipe, "srcx", &src);
+ if (status != ML_ERROR_NONE)
+ goto fail_pipe_destroy;
+
+ status = ml_pipeline_start(pipe);
+ if (status != ML_ERROR_NONE)
+ goto fail_src_release;
+
+ ml_tensors_info_create(&in_info);
+ ml_tensors_info_set_count(in_info, 1);
+ ml_tensors_info_set_tensor_type(in_info, 0, ML_TENSOR_TYPE_FLOAT32);
+ ml_tensors_info_set_tensor_dimension(in_info, 0, in_dim);
+
+ for (int i = 0; i < TOTAL_TEST_SIZE; i++) {
+ std::string path = data_path;
+ path += "testset";
+ printf("\n[%s]\n", path.c_str());
+ std::string img = path + "/";
+ img += "test" + std::to_string(i + 1) + ".bmp";
+ printf("%s\n", img.c_str());
+
+ std::vector<float> featureVector;
+ featureVector.resize(INPUT_SIZE);
+ getInputFeature(tflite_data, img, featureVector);
+
+ status = ml_tensors_data_create(in_info, &in_data);
+ if (status != ML_ERROR_NONE)
+ goto fail_info_release;
+
+ status = ml_tensors_data_get_tensor_data(in_data, 0, &raw_data, &data_size);
+ if (status != ML_ERROR_NONE || data_size != INPUT_SIZE) {
+ ml_tensors_data_destroy(&in_data);
+ goto fail_info_release;
+ }
+
+ for (size_t ds = 0; ds < data_size; ds++) {
+ ((float *)raw_data)[i] = featureVector[i];
+ }
+
+ status = ml_pipeline_src_input_data(src, in_data,
+ ML_PIPELINE_BUF_POLICY_AUTO_FREE);
+ if (status != ML_ERROR_NONE || data_size != INPUT_SIZE) {
+ ml_tensors_data_destroy(&in_data);
+ goto fail_info_release;
+ }
+ }
+
+fail_info_release:
+ ml_tensors_info_destroy(in_info);
+
+ status = ml_pipeline_stop(pipe);
+
+fail_src_release:
+ status = ml_pipeline_src_release_handle(src);
+
+fail_pipe_destroy:
+ status = ml_pipeline_destroy(pipe);
+
+fail_exit:
+ return status;
+#else
+ std::cerr << "Testing of model only with TIZEN" << std::endl;
+ return ML_ERROR_NONE;
+#endif
+}
+
+/**
+ * @brief create NN
+ * Get Feature from tflite & run foword & back propatation
+ * @param[in] arg 1 : configuration file path
+ * @param[in] arg 2 : resource path
+ */
+int main(int argc, char *argv[]) {
+ int status = ML_ERROR_NONE;
+ if (argc < 3) {
+ std::cout << "./TransferLearning Config.ini resources\n";
+ exit(0);
+ }
+
+ const std::vector<std::string> args(argv + 1, argv + argc);
+ std::string config = args[0];
+
+ /** location of resources ( ../../res/ ) */
+ std::string data_path = args[1];
+
+ srand(time(NULL));
+
+ TFLiteData tflite_data;
+ try {
+ setupTensorflowLiteModel(data_path, tflite_data);
+ } catch (...) {
+ std::cerr << "Setting up tflite model failed." << std::endl;
+ return 1;
+ }
+
+ /** Extract features from the pre-trained model */
+ try {
+ extractFeatures(tflite_data, inputVector, labelVector);
+ } catch (...) {
+ std::cerr << "Running tflite model failed." << std::endl;
+ return 1;
+ }
+
+ /** Do the training */
+ status = trainModel(config.c_str());
+ if (status != ML_ERROR_NONE)
+ return 1;
+
+ /** Test the trained model */
+ status = testModel(data_path, tflite_data, config.c_str());
+ if (status != ML_ERROR_NONE)
+ return 1;
+
+ return 0;
}
projects / platform / core / ml / nntrainer.git / commitdiff