6,500 Street Images Segmented for European Telecom AI
Challenge
The client's AI team required pixel-accurate segmentation of outdoor street scenes — classifying surfaces like asphalt, concrete, gravel, pavement bricks, and curbs, as well as objects like poles, street lights, and pedestrians. As a new vendor in their corporate ecosystem, we needed to navigate enterprise procurement approval before any work could begin.
Solution
We worked with the client's team to find a flexible contracting structure that satisfied enterprise requirements, then started with a 500-image pilot batch to calibrate quality standards using CVAT. Three dedicated annotators worked through detailed labeling instructions, achieving approximately 9 minutes per image for full pixel-level segmentation. After the pilot passed quality review, we scaled to production batches of 4,000+ images.
Results
The pilot exceeded accuracy benchmarks, leading to immediate scale-up. The client signed successive batch agreements, growing from 500 to 6,500+ images across three production batches. The client's internal review noted only minor corrections needed, and the partnership continues with new batches in the pipeline.