Overview #
Built a convolutional model to detect vehicles and pedestrians from driving video. The goal was to build a practical detection pipeline and present the model architecture and training decisions to a technical review panel.
Role / Team / Timeline #
Role: Model development and training Team: Inspirit AI (Stanford/MIT alumni led cohort) Timeline: 2023 (2-week program)
Technical Approach #
- Model: VGG-16 based CNN for classification of cars, trucks, jeeps, and pedestrians.
- Data: Custom Kaggle video dataset with labeled classes.
- Pipeline: Preprocessing plus edge/pattern recognition to support robustness.
- Training: 5 hours on Google Colab T4 GPU (about 16 hours total development time).
Results #
- Produced a working detector that identifies moving objects in driving scenarios.
- Presented architecture decisions and supervised learning approach to a peer review panel.
Tools / Stack #
- Python
- TensorFlow / Keras
- Google Colab
Links / Media #
- Demo videos and model artifacts available on request.
