.. _model_training:

Training Perception Models
==========================

Object Detection (YOLOv8)
-------------------------

The notebook :doc:`notebook/kitti_yolo_training.ipynb <../../notebook/kitti_yolo_training>` provides a comprehensive guide for training a YOLOv8 object detection model on the KITTI dataset.

*   **Key Steps in the Notebook:**
    1.  KITTI to YOLO Format Conversion.
    2.  Dataset Splitting.
    3.  YAML Configuration (``kitti.yaml``).
    4.  Hyperparameter Tuning (Optuna).
    5.  Training.
    6.  Evaluation.
    7.  Inference & Visualization.
    8.  ONNX Export.

NLP Chatbot Model
-----------------

Training for the NLP chatbot's intent classification model is integrated within :doc:`NLP/adas_chatbot.py <../../NLP/adas_chatbot>`.

*   When the ``Chatbot`` class is initialized, it checks for ``NLP/models/processed_data_pytorch.pkl`` and ``NLP/models/qa_nn_model.pth``.
*   If missing, ``DataPreprocessor`` and ``ModelTrainer`` are invoked to:
    1.  Process ``moroccan_traffic_code.json``.
    2.  Train the ``NeuralNet``.
    3.  Save processed data and the model.