Preventing the burden of false contraventions by detecting cars models in radar snapshots.
This project acts as a simple showcase of ideas for the AIFlash project (November 2018).
It can also be seen as an overview of possible work when dealing with images, and here specifically car model detection.
As such, this project shows examples of:
- downloading and processing data from an existing open dataset
- using transfer learning to classify cars from labelled data
- using expert-crafted feature extraction and clustering to label unlabelled data
- scrapping web sources to build a custom labelled dataset
Each example is treated in a dedicated Jupyter notebook in the
IAFlash directory.Please note that given the short amount of time available, these notebooks should be considered as examples and not a final product.
A short presentation is also available here.
- improve performances
None.
- literature review
- handling the cars dataset
- try transfer learning
- try 'classical' feature extractors and image clustering
- try web scrapping
- cleanup
- we work on an open car dataset from Stanford.
- other potential sources:
- CompCar dataset (needs to contact the authors).
- we use some of the models from the Keras model zoo (trained on ImageNet).
The code in this project is licensed under GNU General Public License v3.0.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent