Telit Cinterion Enhances deviceWISE AI Visual Inspection with NVIDIA TAO 6.0

Telit Cinterion, an end-to-end IoT solutions enabler, announces the latest enhancement to its deviceWISE^® AI Visual Inspection platform through integration with NVIDIA TAO 6.0. This integration brings low-code AI capabilities to industrial visual inspection, allowing manufacturers to leverage prompt-based segmentation, in-context annotations, and powerful foundation models for tasks such as object pose estimation. Now, quality control teams, engineers and production managers can develop and deploy custom AI models more rapidly, identifying defects and optimizing processes with greater precision across factories. 

Launched in August 2023, deviceWISE AI Visual Inspection delivers a turnkey solution that enables manufacturers in automotive, pharmaceutical, healthcare, energy and other industries to and analyze visual data from their production environments. By uncovering early signs of quality control issues — whether at employee workstations, industrial robots, CNC machines, or other equipment — the platform helps teams take immediate corrective action. This proactive approach reduces rework and unplanned downtime, while boosting product quality, operational efficiency, and on-time delivery across manufacturing lines.

The integration with NVIDIA TAO 6.0 builds on this foundation by incorporating advanced annotation tools, such as prompt segmentation and in-context segmentations, which enable auto-labeling of object detection and segmentation masks using simple text prompts — like “missing bolt on assembly line” or “coating variation on component” — without the need for extensive training or fine-tuning. This reduces annotation costs and effort while improving accuracy for specific industrial objects. Because TAO ships containerized micro‑services with pre‑trained foundation models, it integrates into the existing deviceWISE Visual Inspection pipeline with minimal changes.

Additionally, the integration utilizes foundation models, including FoundationPose for estimating object poses, alongside other models like NV-DINOv2, NV-CLIP, and GroundingDINO. These pretrained models, fine-tuned with minimal custom data via transfer learning, support multi-modal and deliver high-performance inferencing. Deployable on a range of platforms from GPUs to edge devices, NVIDIA TAO 6.0 ensures seamless on-premises operation, maintaining robust security for proprietary data without cloud dependencies.