WPODNet: Build with Torch

Introduction

This repository implements the proposed method from ECCV 2018 paper "License Plate Detection and Recognition in Unconstrained Scenarios" in Torch.

The model in Keras is built by the essay author, see sergiomsilva/alpr-unconstrained.

Example
Annotated
Warp perspective
Confidence	0.9841	0.9945	0.9979

Quick Run

Clone this repository

git clone https://github.com/Pandede/WPODNet-Pytorch.git

Install PyTorch depends on your environment.
Install packages in requirements.txt
```
pip3 install -r requirements.txt
```
Download the pretrained weight wpodnet.pth from here

Predict with an image

python3 predict.py  docs/sample/original/03009.jpg          # The path to the an image
                    # docs/sample/original                  # OR the path to the directory with bulk of images
                    -w weights/wpodnet.pth                  # The path to the weight
                    --save-annotated docs/sample/annotated  # The directory to save the annotated images
                    --save-warped docs/sample/warped        # The directory to save the warped images

Future works

Inference with GPU
Inference with bulk of images
Inference with video
Introduce training procedure
The matrix multiplication seems weird in function postprocess, may improve the computation.

Citation

@inproceedings{silva2018license,
  title={License plate detection and recognition in unconstrained scenarios},
  author={Silva, Sergio Montazzolli and Jung, Cl{\'a}udio Rosito},
  booktitle={Proceedings of the European conference on computer vision (ECCV)},
  pages={580--596},
  year={2018}
}