Capstone 1
Inspect AI
The foundational computer vision deployment focused on high-precision object detection, utilizing bounding boxes to identify the general location and class (e.g., "healthy", "defective") of quality anomalies.
- Architecture: Implementation of DETR (DEtection TRansformer) for anchor-free, end-to-end global context modeling.
- Bounding Box Detection: Identifies the presence of objects in an image and draws a rectangular bounding box around them with a localized class label.
- Precision: Achieved 97.6%+ detection accuracy on complex micro-defects and irregular structural anomalies.
- Velocity: Operates at 27x the speed of traditional manual inspection methodologies.
- Impact: Cut inspection cycle times by 45% during the supervised factory pilot phase.
Capstone 2
DeepInspect AI
An advanced, granular extension of the baseline model, advancing from simple object detection to precise instance segmentation to enable sophisticated Hardware-in-the-Loop (HIL) automation.
- Instance Segmentation: Goes beyond bounding boxes by creating a precise, pixel-level mask for each distinct object instance to exactly outline its shape and boundaries.
- Precise RoI Understanding: Delivers a granular understanding of the object's shape, size, and exact location, crucial for applications requiring detailed environmental interaction.
- Accurate Occlusion Handling: Understands the full shape of partially covered objects to provide a better estimation of the true form compared to traditional detection.
- Advanced Application: Implements HIL AI to deploy an Advanced Component-Aware Multi-Object AI Quality Inspector capable of complex automation logic.