Case Study: HERE Technologies achieves faster, scalable road-sign ground truth for map fine-tuning with Appen’s ML-assisted Video Object Tracking

Learn how HERE Technologies partnered with Appen on their journey to create an understanding of every road sign on earth

HERE Technologies, a global leader in mapping, navigation, and location solutions, needed to build ground-truth data to identify and precisely locate every road sign on earth so their maps and sign-detection models could reach centimeter-level accuracy. Annotating video frame-by-frame was prohibitively slow and expensive, so HERE partnered with Appen and its Machine Learning-assisted Video Object Tracking tool to address the challenge.

Appen delivered a hybrid workflow where a human labels the first video frame and a deep-learning ensemble propagates those labels through subsequent frames while human reviewers correct as needed. This approach drastically speeds annotation compared with frame-by-frame labeling, allowing HERE to scale ground-truth collection across tens of thousands of kilometers, cut annotation time and cost, and produce actionable labeled data far faster for model fine-tuning.

HERE Technologies