How a Top German OEM Increased Assembly Line Efficiency by 4x with Sama

A major German OEM wanted to improve their quality inspection process on the assembly line by implementing cameras instead of humans to detect issues before the vehicles left the factory.

Maintaining quality among complex, changing workflows

To address these issues, the OEM wanted to implement cameras instead of humans on the assembly line to detect quality-related issues before the vehicles left the factory. To help accelerate model training, they turned to Sama’s expertise to quickly and accurately complete 2D video annotations and improve data accuracy.

To ensure consistency and quality, Sama established clear instructions during project kickoff, including tracking frameworks and coordinate conventions. All annotators also underwent a two-week, client-specific training with quality rubrics and golden tasks before tackling the OEM’s data. 

Delivering high-quality, accurate annotations at scale

Our in-house, dedicated team annotated each object of interest (e.g. logo placement, screw placement, series numbers) with bounding boxes and polygons to track objects between frames. Our data experts also reviewed outputs to evaluate the model’s performance, whether it successfully masked irrelevant elements and accurately segmented key parts of each vehicle.

“Training a model to accurately classify elements for a wide variety of complex scenarios is never straightforward,” said Abha Laddha, Sr. Director, Solutions & Launch. “For this project, we needed to consider a number of quality issues that could arise, from parts defects to surface imperfections like scratches or dust. Our early edge case detection and flexible workflows helped us proactively create new training instructions and get it right the first time. Plus, regular collaborative sessions with the OEM and our integrated calibration process meant that our teams were aligned throughout the entirety of the project.”

‍

Workflows

Sama teams annotated each object of interest (e.g. logo placement, screw placement, series numbers) with bounding boxes and polygons to track objects between frames.

They then evaluated outputs to ensure irrelevant elements were successfully masked and key parts of each vehicle were accurately segmented.

‍

Results

Sama achieved a 99% acceptance rate and helped the OEM:

• Increase assembly line efficiency by 4x
• Decrease rework and errors on vehicle orders
• Reduce total cost of ownership by eliminating project delays and revisions

‍

_{Photo credit}