Inline quality inspection
Micro-defect, assembly completeness, and packaging verification with tunable thresholds per SKU.
Computer Vision
Perception systems for the physical operating world.
Custom CNNs and Vision Transformer models trained on your domain data for real-time detection, anomaly finding, and classification, including edge deployments when data cannot leave the site.
Perspective
Vision that survives the factory floor
Computer Vision pilots often fail in production because training data never captured glare, motion blur, line changeovers, or the long tail of defect morphology.
InheritX builds perception pipelines with field validation loops: capture protocols, labeling governance, and retraining triggers tied to operator feedback, not static model drops.
We design for deployment reality, GPU budgets at the edge, latency budgets on high-speed lines, and integration paths into quality hold systems and maintenance ticketing.
Edge-ready
On-line inference within SLA
Explainable
Heatmaps & review UI for QA
Connected
MES, SCADA, CMMS hooks
Maintainable
Retrain paths when lines shift
How we engage
Vision system lifecycle
From feasibility through sustained operation on live lines.
01
Feasibility & optics
Camera placement, lighting, and frame-rate study against target defect classes and line speed constraints.
02
Labeled corpus build
Structured capture across shifts and SKUs with inter-annotator agreement checks and versioned datasets.
03
Model & edge packaging
Train, quantize, and benchmark on target hardware; define fallback behavior when inference confidence drops.
04
Line integration
Reject mechanisms, HMI alerts, and data feeds to quality dashboards, coordinated with controls engineers.
05
Monitoring & refresh
Track precision/recall drift, sample false positives for relabeling, and schedule retrains after material or tooling changes.
Capabilities
Use cases we deliver
Safety & compliance monitoring
PPE detection, restricted-zone intrusion, and procedural adherence alerts with clip retention policies.
Inventory & yard awareness
Slot occupancy, damage assessment, and asset identification fused with WMS events.
Document and label capture
High-speed OCR and barcode validation on inbound logistics streams with exception routing.
Capabilities
Edge and on-device AI
When frames cannot leave the line, we package perception to run next to the camera, not only in a cloud GPU pool.
On-device / line-side inference
Latency-bounded models on edge GPUs or industrial PCs, with fallback and hold logic when confidence drops.
Model packaging for the floor
Export and optimize for the hardware you already run (ONNX / TensorRT-class paths) inside your patch and rollback process.
Labeling and retraining loops
Operator overrides feed a governed labeling queue, synthetic or sampled data only where it improves the long tail, not as a standalone product.
Vision stack we actually use
PyTorch training, OpenCV capture pipelines, and detector families such as YOLO or ViT, chosen for the defect class and SLA, not a logo wall.
Dual view
Edge vs. cloud placement
Edge inference
Sub-second decisions on high-throughput lines
Bandwidth-constrained or air-gapped sites
Deterministic actuation tied to PLCs
Cloud aggregation
- Cross-site model training and benchmarking
- Long-horizon analytics on inspection trends
- Centralized retraining pipelines with federated upload
- Executive dashboards on quality and downtime drivers
FAQ
Computer Vision FAQ
We specify requirements and work with your preferred vendors; integration and model delivery are our core scope.
Review UIs show evidence overlays, support quick override, and feed corrections back into the training queue.
Yes, we often augment legacy inspection with AI on defect classes rules miss, sharing outputs into the same quality workflow.
No. Edge is a deployment topology for the same vision system, used when data residency, bandwidth, or line latency requires on-device inference.
Continue
Related offerings
Next step
Map this capability to your mandate.
A focused strategy conversation, constraints, systems, and what production readiness looks like for your organization.