README.md

ANPR-GIA

Set up

1. Create a virtual environment

conda env create -f environment.yml
conda activate yolo

2. Install pytorch
   For macOS:
   conda install pytorch::pytorch torchvision -c pytorch
   For Windows or Linux, if you have Nvidia GPU:
   conda install pytorch torchvision pytorch-cuda=11.8 -c pytorch -c nvidia
   For Linux, if you have AMD GPU:
   pip install torch torchvision --index-url https://download.pytorch.org/whl/rocm6.0
   If not, install for CPU:
   conda install pytorch torchvision cpuonly -c pytorch

Repo structure

• pipeline.py: contains the full and clean implementation of our pipeline to detect and recognize license plates.
• notebooks/: folder with notebooks of the process that lead to the final code.
  • comparisonOCR.ipynb: script with metrics comparing Tesseract, EasyOCR and PaddleOCR for recognition.
  • comparisonYoloMathMorph.ipynb: script with examples and metrics of the two detection models we used.
  • customOCR.ipynb: an attempt to train a Convolutional Neural Net on EMNIST
  • evaluation.ipynb: an example of use of our pipeline, along with its evaluation on the test set.
  • generate_dataset_recognition.ipynb: script to generate a syntetic dataset of license plates to recognize.
  • mathMorph.ipynb: the process followed to implement a mathematical morphology method to detect license plates.
  • segmentation.ipynb: evaluation of only the segmentation step of our pipeline.
  • yolo11.ipynb: training of the YOLOv11 detection model and example inference.
• Models/: folder containing the trained yolo11n_licenseplates.pt, which can also be found on HuggingFace.