NEW DELHI, June — SunTec India completed a major data labeling initiative for a government-backed infrastructure consulting firm, processing more than 3 million annotations with sustained 99% accuracy. The dataset will train a highway asset & damage detection model across India’s national and state highway networks.
Operating on a client-hosted CVAT platform, the team labeled highway images using bounding boxes and 4-point/multi-point polygon annotation to identify road surface damage and classify road assets in accordance with IRC82 government standards, covering 71 distinct damage and asset categories across more than 1,000 kilometers of highway corridor. 4-point annotations made up approximately 80% of the total workload, while the remaining 20% required highly detailed multi-point annotations.
The image annotation engagement began as a pilot with 4 annotators and 1 reviewer. After a successful pilot, the scope expanded sharply, and the team scaled to 35 annotators and 7 reviewers, all of whom were subject-matter experts with backgrounds in Civil Engineering. The accuracy rate held through that eightfold growth, which the company credits to a deliberate reviewer-to-annotator ratio and structured training on IRC82 standards before labeling live data.
“What I’m proudest of is that the accuracy held as the scale grew,” said Rohit Bhateja, Director of Digital Engineering Services and Head of Marketing at SunTec India. He added, “A highway isn’t uniform—weather, lighting, and wear differ by region. Labeling 71 categories across a thousand kilometers throws up thousands of edge cases, and hitting 99% accuracy there proves our human-in-the-loop workflow holds up regardless of dataset size.”
The milestone reinforces SunTec India’s position in domain-intensive data labeling for AI and machine learning applications, particularly geospatial and aerial image labeling. It also reflects growing demand for specialized AI training data, with technical annotation guidelines and dense categories. The company expects to see more demand for this kind of high-consistency labeling across regulated industries.
