Edge AI
Edge inference constraints — quantization, NPU throughput, and interface bandwidth for on-device vision AI.
-
AOI Only Says “Defective” — How Vision Language Models Fill the Gap in Judgment Rationale
AOI systems report only pass or fail, never why. This article examines how vision-language models can supply the missing judgment rationale, turning an inspector's tacit expertise into structured, auditable evidence.
-
Camera Interface Bandwidth Determines Resolution — From GigE Vision to CoaXPress 2.0
Six 12MP cameras on a PCB inspection line hit a cable bottleneck before lens or lighting even matter. This article maps how GigE Vision versus CoaXPress 2.0, and edge AI on-device inference, trade…
-
Few-Shot Anomaly Detection: Can 5 Defect Samples Stand Up an Inspection Line
A question repeats every time a new line is set up: how many defect samples before an inspection model can run? Few-shot anomaly detection uses mostly normal data plus 5-20 defect samples to…
-
What You Must Check Before Applying INT8 Quantization to Edge AI Inference: Why Micro-Defects Quietly Disappear
A micro-scratch detection model quantized for edge inference looks fine on overall mAP, but a specific low-contrast defect class can quietly lose accuracy. Layer-wise quantization, edge hardware, and camera interface all interlock in…
-
Detection Time Cut 65%, False Alarms Down 80%: How to Read the Two Numbers Behind Edge Fire-Detection AI
A machine-vision read on an early-detection model for underground-garage EV fires running in real time on an edge device. Of the two headline figures, the one that matters more in the field is…
-
Automotive Part Appearance Inspection: Lighting Design Comes First
Automotive parts mix painted, chrome-plated, and molded surfaces with very different reflectivity on one line. Why lighting design matched to each material has to be settled before the AI algorithm can do its…
-
In the Age of Deep-Learning Inspection, What to Check Before Camera Spec
Hardware suppliers are quietly building out software portfolios for AI inspection. Why deep-learning classification only works on top of an optical setup that already delivers the defect in the image — not instead…
-
Edge AI Inference Latency: The Camera Interface Decides Before the NPU Does — GigE Vision 3.0 RDMA and Disaggregated Inspection Architecture
A new NPU didn’t fix your throughput problem? Trace the bottleneck and it’s often the camera-to-host interface, not compute — and GigE Vision 3.0 RDMA (GVRSP) is the fix.
-
What Should a Camera See When a Humanoid Grasps an Object?
Humanoid grasping accuracy hinges on the vision/edge-AI pipeline -- a look at the on-device vs. cloud-offload trade-off and the INT8 quantization-loss problem.
-
What Changes in Vision Inspection When Logistics Automation Moves Into the Front End
As front-end semiconductor logistics automation expands, vision-based position checkpoints at transport handoffs become the standard defense against alignment drift.
