Object Detection and Tracking
We handle object continuity with a broad range of object detection models, not only YOLO-family approaches, together with MOT-based tracking.
Technology
Vision AI
To extract meaningful events from video, we work not only on models but also on preprocessing, tracking, and post-processing.
Vision AI cannot be explained by model accuracy alone. Camera placement, resolution, target size, lighting, occlusion, and motion all change how the full pipeline should be designed.
Odette combines object detection and tracking, segmentation, and computer-vision-based inspection according to the field problem. We also handle vision systems tied to operations, such as access control, attendance, face recognition, and CCTV analytics.
The whole pipeline matters more than any single model.
Model architecture, datasets, preprocessing, and post-processing all change depending on whether the task is to detect small objects, follow fast-moving targets, or segment regions precisely. Odette designs the pipeline only after understanding that problem profile in detail.
Even within object detection, we do not rely only on YOLO-family models. We have worked across a broad range of detection approaches and can design and implement architectures directly when the problem requires it.
For inspection tasks, we use not only deep-learning approaches such as PatchCore, but also machine-learning and rule-based methods, choosing the most practical approach for the data and operating constraints.
Technical Scope
Segmentation
We build deep-learning segmentation pipelines for tasks that require pixel-level segmentation and region-based analysis.
Defect Inspection
We handle defect inspection with both deep-learning approaches such as PatchCore and machine-learning-based rule systems.
Preprocessing and Correction
We refine input quality through distortion correction, ROI handling, and image enhancement.
Operational Vision Systems
We extend vision outputs into operational systems such as access control, attendance, face recognition, and CCTV monitoring.
Data Collection and Labeling
We have practical experience adapting labeling tools built on CVAT and PPOCRLabel, and can structure data collection plans, labeling criteria, and review workflows based on real project work.