This project was something I was really passionate about. I started working on it with the goal of building a complete and functional model, but unfortunately, it failed due to the lack of proper data. I spent quite some time trying to collect and prepare the right dataset to make it work, but things didn’t go as planned.
slow_traffic_small.mp4
I made several modifications to this video to improve its usability and see if I could extract meaningful results from it.
🚗 Data Challenges
The lack of proper traffic video datasets with Persian license plates pushed us to try a lot of different approaches. We experimented with custom video collection, synthetic data generation, frame extraction, and even manual labeling — just to make the model understand Persian plates better. It was a tough process, but it helped us realize how critical real-world localized datasets are for accurate model performance.
Still, it was a great learning experience — I learned a lot about data handling, model optimization, and the real challenges behind working with limited datasets. Maybe someday I’ll revisit this project once I find or create a better dataset.