return APP_ERROR_NONE;
}
-static void on_data_receive_(ml_tensors_data_h data,
- const ml_tensors_info_h info, void *user_data) {
+int util_save_drawing(cairo_surface_t *cr_surface, const char *dst) {
+ cairo_status_t cr_stat = CAIRO_STATUS_SUCCESS;
+
+ LOG_D("start writing to png_path: %s ", dst);
+ /// @todo change to other format
+ cr_stat = cairo_surface_write_to_png(cr_surface, dst);
+
+ if (cr_stat != CAIRO_STATUS_SUCCESS) {
+ LOG_E("failed to write cairo surface as a file reason: %d", cr_stat);
+ return APP_ERROR_INVALID_PARAMETER;
+ }
+
+ return APP_ERROR_NONE;
+}
+
+static void on_feature_receive_(ml_tensors_data_h data,
+ const ml_tensors_info_h info, void *user_data) {
appdata_s *ad = (appdata_s *)user_data;
void *raw_data;
pthread_mutex_unlock(&ad->pipe_lock);
}
-static int run_nnpipeline_(appdata_s *ad, const char *src) {
+static int run_mobilnet_pipeline_(appdata_s *ad, const char *src) {
char pipe_description[5000];
-
char model_path[PATH_MAX];
int status = ML_ERROR_NONE;
util_get_resource_path("mobilenetv2.tflite", model_path, false);
- status = pthread_mutex_lock(&ad->pipe_lock);
- if (status != 0) {
- LOG_E("acquiring lock failed status: %d", status);
- return status;
- }
-
LOG_D("pipe ready, starting pipeline");
sprintf(pipe_description,
LOG_D("setting inference \n pipe: %s", pipe_description);
status = ml_pipeline_construct(pipe_description, NULL, NULL, &ad->pipeline);
-
if (status != ML_ERROR_NONE) {
LOG_E("something wrong constructing pipeline %d", status);
- ml_pipeline_destroy(ad->pipeline);
- pthread_mutex_unlock(&ad->pipe_lock);
return status;
}
- status = ml_pipeline_sink_register(ad->pipeline, "sink", on_data_receive_,
+ status = ml_pipeline_sink_register(ad->pipeline, "sink", on_feature_receive_,
(void *)ad, &ad->pipe_sink);
if (status != ML_ERROR_NONE) {
LOG_E("sink register failed %d", status);
- goto CLEAN;
+ goto PIPE_DESTORY;
}
LOG_D("starting inference");
status = ml_pipeline_start(ad->pipeline);
if (status != ML_ERROR_NONE) {
LOG_E("failed to start pipeline %d", status);
- goto CLEAN;
+ goto SINK_UNREGISTER;
+ }
+
+ status = pthread_mutex_lock(&ad->pipe_lock);
+ if (status != 0) {
+ LOG_E("acquiring lock failed status: %d", status);
+ goto MUTEX_UNLOCK;
}
pthread_cond_wait(&ad->pipe_cond, &ad->pipe_lock);
status = ml_pipeline_stop(ad->pipeline);
if (status != ML_ERROR_NONE) {
LOG_E("stopping pipeline failed");
- goto CLEAN;
+ goto MUTEX_UNLOCK;
}
- LOG_D("unregister pipeline");
- if (status != ML_ERROR_NONE) {
- LOG_E("unregistering sink failed");
- }
+MUTEX_UNLOCK:
+ pthread_mutex_unlock(&ad->pipe_lock);
-CLEAN:
- LOG_D("destroying pipeline");
+SINK_UNREGISTER:
ml_pipeline_sink_unregister(ad->pipe_sink);
- ml_pipeline_destroy(ad->pipeline);
ad->pipe_sink = NULL;
+
+PIPE_DESTORY:
+ ml_pipeline_destroy(ad->pipeline);
ad->pipeline = NULL;
- pthread_mutex_unlock(&ad->pipe_lock);
+
return status;
}
char png_path[PATH_MAX];
const char *dst =
ad->tries < MAX_TRAIN_TRIES ? TRAIN_SET_PATH : VALIDATION_SET_PATH;
- cairo_status_t cr_stat = CAIRO_STATUS_SUCCESS;
int status = APP_ERROR_NONE;
- util_get_data_path("temp.png", png_path);
- LOG_D("start writing to png_path: %s ", png_path);
- cr_stat = cairo_surface_write_to_png(ad->cr_surface, png_path);
+ status = util_get_data_path("test.png", png_path);
+ if (status != APP_ERROR_NONE) {
+ LOG_E("getting data path failed");
+ return status;
+ }
- if (cr_stat != CAIRO_STATUS_SUCCESS) {
- LOG_E("failed to write cairo surface as a file reason: %d", cr_stat);
- return APP_ERROR_INVALID_PARAMETER;
+ status = util_save_drawing(ad->cr_surface, png_path);
+ if (status != APP_ERROR_NONE) {
+ LOG_E("failed to save drawing to a file");
+ return status;
}
util_get_data_path(dst, ad->pipe_dst);
LOG_I("start inference to dataset: %s ", ad->pipe_dst);
- status = run_nnpipeline_(ad, png_path);
+ status = run_mobilnet_pipeline_(ad, png_path);
return status;
}
LOG_E("pipe write error");
return NULL;
};
- usleep(150);
}
LOG_D("training finished");
return status;
}
+
+static void on_inference_end_(ml_tensors_data_h data,
+ const ml_tensors_info_h info, void *user_data) {
+ appdata_s *ad = (appdata_s *)user_data;
+
+ float *raw_data;
+ size_t data_size;
+
+ int status = pthread_mutex_lock(&ad->pipe_lock);
+ if (status != 0) {
+ LOG_E("acquiring lock failed %d", status);
+ pthread_cond_signal(&ad->pipe_cond);
+ return;
+ }
+
+ status =
+ ml_tensors_data_get_tensor_data(data, 0, (void **)&raw_data, &data_size);
+ if (status != ML_ERROR_NONE) {
+ LOG_E("get tensor data failed: reason %s", strerror(status));
+ goto RESUME;
+ }
+
+ if (data_size != sizeof(float) * NUM_CLASS) {
+ LOG_E("output tensor size mismatch, %d", (int)data_size);
+ goto RESUME;
+ }
+
+ /// SMILE: 0 1
+ /// FROWN: 1 0
+ LOG_D("label: %lf %lf", raw_data[0], raw_data[1]);
+
+RESUME:
+ status = pthread_cond_signal(&ad->pipe_cond);
+ if (status != 0) {
+ LOG_E("cond signal failed %d", status);
+ }
+ pthread_mutex_unlock(&ad->pipe_lock);
+}
+
+int run_inference_pipeline_(appdata_s *ad, const char *filesrc) {
+ char pipe_description[9000];
+ char tf_model_path[PATH_MAX];
+ char trainer_model_path[PATH_MAX];
+
+ int status = APP_ERROR_NONE;
+
+ status = util_get_resource_path("mobilenetv2.tflite", tf_model_path, false);
+ if (status != APP_ERROR_NONE) {
+ LOG_E("error getting resource path, reason: %d", status);
+ return status;
+ }
+
+ status = util_get_resource_path("model.ini", trainer_model_path, false);
+ if (status != APP_ERROR_NONE) {
+ LOG_E("error getting data path, reason: %d", status);
+ return status;
+ }
+
+ sprintf(pipe_description,
+ "filesrc location=%s ! pngdec ! videoconvert ! "
+ "videoscale ! video/x-raw,width=224,height=224,format=RGB ! "
+ "tensor_converter ! "
+ "tensor_transform mode=arithmetic option=%s ! "
+ "tensor_filter framework=tensorflow-lite model=%s ! "
+ "tensor_filter framework=nntrainer model=%s input=1280:7:7:1 "
+ "inputtype=float32 output=1:%d:1:1 outputtype=float32 ! "
+ "tensor_sink name=sink",
+ filesrc, "typecast:float32,add:-127.5,div:127.5", tf_model_path,
+ trainer_model_path, NUM_CLASS);
+
+ LOG_D("pipe description: %s", pipe_description);
+ status = ml_pipeline_construct(pipe_description, NULL, NULL, &ad->pipeline);
+ if (status != ML_ERROR_NONE) {
+ LOG_E("constructing pipeline failed, reason: %d", status);
+ goto PIPE_UNLOCK;
+ }
+
+ status = ml_pipeline_sink_register(ad->pipeline, "sink", on_inference_end_,
+ (void *)ad, &ad->pipe_sink);
+ if (status != ML_ERROR_NONE) {
+ LOG_E("sink register failed, reason: %d", status);
+ goto DESTORY_PIPE;
+ }
+
+ status = ml_pipeline_start(ad->pipeline);
+ if (status != ML_ERROR_NONE) {
+ LOG_E("failed to start pipeline %d", status);
+ goto UNREGISTER_SINK;
+ }
+
+ status = pthread_mutex_lock(&ad->pipe_lock);
+ if (status != 0) {
+ LOG_E("acquiring lock failed status: %d", status);
+ goto UNREGISTER_SINK;
+ }
+
+ pthread_cond_wait(&ad->pipe_cond, &ad->pipe_lock);
+
+ status = ml_pipeline_stop(ad->pipeline);
+ if (status != ML_ERROR_NONE) {
+ LOG_E("stopping pipeline failed");
+ }
+
+PIPE_UNLOCK:
+ pthread_mutex_unlock(&ad->pipe_lock);
+
+UNREGISTER_SINK:
+ ml_pipeline_sink_unregister(ad->pipe_sink);
+ ad->pipe_sink = NULL;
+
+DESTORY_PIPE:
+ ml_pipeline_destroy(ad->pipeline);
+ ad->pipeline = NULL;
+
+ return status;
+}
+
+int data_run_inference(appdata_s *ad) {
+ char png_path[PATH_MAX];
+
+ int status = APP_ERROR_NONE;
+
+ status = util_get_data_path("test.png", png_path);
+ if (status != APP_ERROR_NONE) {
+ LOG_E("getting data path failed");
+ return status;
+ }
+
+ status = util_save_drawing(ad->cr_surface, png_path);
+ if (status != APP_ERROR_NONE) {
+ LOG_E("saving the cairo drawing failed");
+ return status;
+ }
+
+ status = run_inference_pipeline_(ad, png_path);
+
+ return status;
+}
projects / platform / core / ml / nntrainer.git / commitdiff