The wildlife detection system I built for railway tracks at IIRS-ISRO taught me something that…

no course ever did.

In the lab, my model detected animals at 96% accuracy. On actual railway footage? It dropped to 71%. Rain, fog, nighttime, motion blur — real-world conditions are brutal.

I spent three months not improving the model architecture, but improving the data pipeline. Better augmentation for weather conditions. Synthetic nighttime images. Edge cases from actual field footage.

Final accuracy in production: 93%.

The takeaway that stuck with me: production computer vision is 20% model and 80% engineering. Data pipeline quality, edge deployment optimization, handling real-world conditions — that's where the actual work lives.

If you're in CV, stop chasing the latest architecture paper and start obsessing over your data quality and deployment pipeline. That's what makes the difference between a demo and a system that actually works on a rainy Tuesday night.