NexVDO SDK · Module A/04
Analysis
Your models, against live video, on-device. Run detection, segmentation, recognition, and behavior analysis against live video. Deploy on cloud or edge systems with NVIDIA and Intel GPUs.
- Content · biometric · behavior
- NVIDIA · Intel GPUs
- Cloud · edge
Model to decision
Keep live video and analysis in one deployment path.
The Analysis module applies content, biometric, and behavior functions to live video. YUAN publishes cloud and edge deployment with NVIDIA and Intel GPU integration.
YUAN announced Everything Detection, Everything Segmentation, and Image Description in March 2024. The documented QDEEP custom-weights path exists.
frame = live_video_frame
model = custom_weights_path
result = analyze(frame, model)
output = detection | segmentation | counting | text_recognition
deployment = edge | cloud// Illustrative sketch — not a published API
- 01
Live video
- 02
Custom weights
- 03
Analysis
- 04
Decision
Illustrative Analysis pipeline from live video to a decision.
Analysis capabilities
Three analysis blocks. One live-video path.
Detect, segment, count, and match
Run object detection, image segmentation, classification counting, and feature matching. Read license plates and other text in the frame.
Analyze faces from more than one angle
Run face recognition and multi-angle face analysis. Classify age, gender, and expression.
Turn skeleton data into event states
Use a 17-keypoint skeleton for posture and movement analysis. Detect falling, wandering, and throwing events across continuous video.
Published surface
The capability blocks are published. Runtime details are not.
YUAN publishes the analysis functions, deployment locations, and GPU families. Model formats and custom-weights mechanics remain unpublished.
| Capability block | Published functions | Compute and delivery |
|---|---|---|
| Content analysis | Detection · segmentation · counting · feature matching · license-plate and text recognition | Cloud · edge · NVIDIA and Intel GPUs |
| Biometric analysis | Face recognition · multi-angle · age · gender · expression | Cloud · edge · NVIDIA and Intel GPUs |
| Behavior analysis | 17-keypoint skeleton · fall · wander · throw detection | Cloud · edge · NVIDIA and Intel GPUs |
| Generative AI features | Everything Detection · Everything Segmentation · Image Description | Feature-to-platform mapping — pending publication |
| Custom weights path | QDEEP path documented | QDEEP mechanics — pending publication |
| Model formats | Accepted model formats — pending publication | Runtime mapping — pending publication |
Analysis capability blocks × published functions × compute and delivery surface.
Evaluation artifacts
Proof arrives with a named state.
Nothing is represented as public until it is public.
- DatasheetDatasheet — pending publication
- Sample codeSample code — published with evaluation access
- Compatibility matrixCompatibility matrix — pending publication
Next module
Continue through the SDK.
Buying a product?
You are building your own device — NexVDO SDK plus YUAN hardware inside your product.
Start at the SDK01Commissioning a system?
You need an integrated stack delivered and supported — hardware, software, and integration through Neteon.
Request scoping02Need a human?
Engineers answer evaluation requests within 48 hours. Business and hardware questions route to Neteon — YUAN’s U.S. front door.
Run your model on the target edge system
Included free with YUAN hardware, delivered through a supported evaluation path — engineer answer within 48 hours. Share the video source, target hardware, and model goal. An engineer answers with scope.
- Engineer reply in 48h
- Ships from U.S. stock
- Free with YUAN hardware