Project overview

OpenSpot (FindSpot) is a computer vision project whose goal is to detect and classify parking spaces from images taken from an elevated or aerial point of view. The project focuses not only on prediction accuracy, but also on deployment constraints such as model size and inference speed.

Problem statement

In real-world applications, parking availability information is often incomplete or outdated. An automated vision-based approach can help transform raw visual data into actionable information in real time.

Models evaluated

• MobileNetV3-Small – lightweight and deployment-friendly
• EfficientNet-B0 – strong accuracy–efficiency trade-off
• ResNet18 – robust baseline with very low latency

Results summary

The models were compared using accuracy, inference speed (FPS and latency), and model size. MobileNetV3-Small achieved the best overall balance between accuracy and efficiency, making it the most suitable choice for a web or mobile deployment scenario.

Conclusion

OpenSpot illustrates the importance of evaluating machine learning models beyond raw accuracy. By considering deployment constraints early, the project bridges the gap between academic experimentation and real-world applications.