https://github.com/author31/license_plate_detection
This repository aims to re-implement a well-known object detection method FasterRCNN using fastai.
https://github.com/author31/license_plate_detection
deep-learning fastai nbdev object-detection plate-detection re-implementation
Last synced: 3 days ago
JSON representation
This repository aims to re-implement a well-known object detection method FasterRCNN using fastai.
- Host: GitHub
- URL: https://github.com/author31/license_plate_detection
- Owner: author31
- License: apache-2.0
- Created: 2022-10-04T11:01:52.000Z (over 3 years ago)
- Default Branch: main
- Last Pushed: 2022-10-15T08:14:25.000Z (over 3 years ago)
- Last Synced: 2023-10-27T16:32:33.931Z (over 2 years ago)
- Topics: deep-learning, fastai, nbdev, object-detection, plate-detection, re-implementation
- Language: Jupyter Notebook
- Homepage:
- Size: 8.8 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
License Plate Detection
================
## Install
`pip install license_plate_detection`
## How to use
### Download the dataset
Download the dataset from https://github.com/detectRecog/CCPD and put
all of that into a directory. There is a ./exp/license_ds.csv file
contains only samples of ccpd_base directory (there are few other of
ccpd\_\* for other weather conditions) which is for faster data loading
in fastai. For simplicity, I only used ccpd_base.
### Experimenting
- There is a notebook named /exps/faster_rcnn.ipynb that illustrates a
whole training neural workflow from loading dataset to inference
### TODO
- Model haven’t been well trained, there must be some wrong
implemtations
- Improve NMS operation, the current stage of region refinement of RPN
is lack of NMS because the pure NMS is slow which may affects the
speed of training
- Add OCR feature for detecting the license plate number of each
predictions
- Add RPN config for easier customization