Major project
Automotive, EV & Aerospace
Road scene understanding for Indian roads
Segment roads, vehicles and people in driving images from Indian traffic.
The brief
Driver-assistance systems trained on European roads struggle with Indian traffic: auto-rickshaws, animals, unmarked lanes. Train a segmentation model on Indian driving images, compare it with one trained on a Western dataset, and show where each fails. Good looks like: a measured gain on Indian scenes, examples of hard cases, and a model that runs on a modest GPU.
What you will build
- Prepare the Indian driving images and labels
- Train a semantic-segmentation model
- Compare with a model trained on a Western dataset
- Analyse failures by class and scene type
- Optimise for speed on a modest GPU
- Demo on a short video and write the report
Free, open data to build it with
- India Driving Dataset (IDD) ↗ IIIT Hyderabad
- KITTI Vision Benchmark ↗ KIT and Toyota Technological Institute
- comma2k19 driving data ↗ comma.ai (GitHub)
Check each dataset's licence before you publish your work. Your mentor gives your team its own version of the brief (data, scope or region), so every team's project is its own.