SOD C/C++ API - RealNet Architecture


int sod_realnet_train_start(sod_realnet_trainer *pTrainer, const char *zConf);


Start the RealNet training process. Depending on your dataset, training should take at least 18 hours to complete on a modern CPU. Once done, a RealNet model should be saved on disk. Besides preparing your dataset (positive, negative and test samples) for example for a standard object detection purpose, you should at least register a log consumer callback and a path where to store the RealNet output model before starting the training phase. All of this is done via sod_realnet_train_config().


sod_realnet_trainer    *pTrainer

A pointer to a valid sod_realnet_trainer object obtained from a prior successful call to sod_realnet_train_init().

const char    *zConf

A pointer to a null terminated memory buffer or a file on disk holding the training instructions (i.e. where the dataset is located, tree minimal depth and so forth).

A typical training instructions template (i.e. skeleton) should look like the following. Feel free to adjust, tune any parameter you want for your specific needs.

Return Value

SOD_OK is returned on success. Any other code indicates failure.


Checkout the introduction course, the C/C++ samples on the download page or refer to the SOD Github Repository.

See also