Sharper, Faster, Smarter: Focoos AI Vision running on Lenovo Power
How Focoos and Lenovo are Transforming Computer Vision in Manufacturing
Lenovo
Screw Detection
Cable Damage Detection
Advanced Manufacturing in Industrial Automation
Custom AI model development
Real-time computer vision deployment
Platform-based model management
Precise detection for manufacturing quality and operational safety.
Objective
Lenovo is a global technology leader known for its innovative hardware solutions, including high-performance computing systems used in enterprise and AI applications. The collaboration between Focoos AI and Lenovo focused on validating the effectiveness of Focoos computer vision models for real-world industrial quality control tasks. Two high-impact use cases were selected: screw detection, essential for verifying assembly integrity, and cable damage detection, crucial for safety and reliability in electronics manufacturing. Each task was tested under realistic conditions using Lenovo ThinkSystem SR670 V2 servers with NVIDIA L40S GPUs. Models were benchmarked against industry-standard solutions like YOLOv8 and RT-DETR-v2 using mAP and FPS as performance metrics.
Screw Detection
In this task, precision was key to identifying small and partially occluded components on high-speed production lines. Focoos AI’s large model reached a 78.8 AP50, the highest among all evaluated models, while maintaining over 400 FPS. The small variant further pushed the limits with 655 FPS and only a slight trade-off in precision.rnCompared to YOLOv8-L, which topped at 75.1 AP50 and dropped to just 163 FPS, Focoos models delivered both better accuracy and up to 4x faster inference, making them ideal for inline quality checks where latency cannot be tolerated. This balance of power and speed is visualized in the chart, where all Focoos variants outperform competitors in the upper-right quadrant.
Cable Damage Detection
Unlike screw detection, cable damage tasks require high sensitivity to subtle anomalies, such as frayed shielding or small tears, often in varying lighting or overlapping objects. Focoos AI’s large and small models consistently surpassed competitors, achieving up to 90.5 AP50 at 655 FPS.rnNotably, even top-performing alternatives like YOLOv8-S and RT-DETR-v2-M plateaued at 90.2 AP50, but with lower FPS (420 and 410, respectively). The Focoos medium model also stood out by offering a better speed-to-accuracy ratio than most baselines, placing all Focoos models clearly ahead in the upper-right quadrant of the performance chart.rnThis shows that Focoos AI not only detects visible damage but also does so fast enough to keep pace with demanding production environments, a crucial capability when each second of delay can mean costly downtime or missed defects.
+55%
Faster
Real-time inference at up to 655 FPS, enabling instant defect detection on the line.
+5%
Sharper
Higher accuracy on average across use cases, ensuring fewer missed defects.
90%
Faster-to-Deploy
Drastically reduced development time thanks to Focoos AI automated platform.