Damage segmentation for vehicle-perception models.

Challenge

Autonomous-vehicle perception models need enormous volumes of image, video and sensor data, labeled to one consistent standard across thousands of hours of footage. The hard part isn't the easy frames; it's the defect classes, occlusions and rare edge cases, where annotator drift quietly poisons a model.

Approach

io-ai stood up a dedicated, trained team running a strict SOP: tight polygon segmentation of damage and defects across dozens of classes, with explicit rules for partial occlusion, ambiguous boundaries and generic-class fallbacks.

A second workstream handled object detection and tracking (boxes and class labels held consistent across video with persistent track IDs), exported directly to the program's ontology and format.

Quality

Every task ran through io-ai's multi-level QA pipeline, with first-batch audits before scale-up and full rejection-reason metadata on every item, holding over 98% accuracy on rolling audits, with weekly status reporting.

Result

A consistent, audit-traceable perception dataset that fed vehicle-perception models to a safety-critical bar, accurate where it matters most, on the defect classes and edge cases that break weaker pipelines.

This was a scaled engagement run with calibration sessions and zoom-to-confirm boundary checks on every contentious frame.