Car Detection

Car Detection ML pipeline

Description

This project composes a deep learning pipeline for multi-car detection in a given picture. It emphasises the understanding of the underlying detection model - FasterRCNN.

Data

Data used for training and evaluation are not public, but were infered from public City Scapes dataset.

Data Exploration

Data exploration is performed in src/notebooks/data_exploration.

Input Sample

Architecture

We use FasterRCNN neural network as the detection algorithm. It is fully accessible from the torchivsion library, however, we developed it ourselves from sufficiently fundamental building blocks for better understanding.

Inference Example

Usage

In the src folder

Before anything

Make sure to have uv package manager installed on your device. To install necessary libaries to your environment, run:

uv sync

Training

uv run python training.py <dataset_path>

Inference

uv run python inference.py <dataset_path> model.pt 

Evaluation

uv run python evaluation.py DET <dataset_path_bboxes> .\output_predictions\