
Computer Vision Research
Railway Semantic Segmentation Research
- Problem
- Railway inspection workflows need reliable pixel-level understanding of track, ballast, and surrounding scene context.
- Contribution
- Fine-tuned segmentation models on RailSem19-style data, added transfer-learning support in an MMSegmentation fork, and evaluated railway-specific failure modes.
- Impact
- Turned academic segmentation work into a practical inspection pipeline for detecting track-adjacent issues such as mud pumping indicators.

