Release 2025/12/16
v3.7.2
Observ Dec Updates
This release focuses on stepwise VLM workflows, richer scene events, data quality controls, and smoother batch execution—plus a sizable upgrade to our general model label set.
New Features:
✅ VLM Flow
A flexible pipeline that brings together Object Detection, Time Gap, and VLM in one logical flow. It filters frames using efficient low-cost signals first, then calls the VLM only when needed—delivering smarter, more accurate results with less compute.
Stepwise reasoning: Support coarse → fine stages (e.g., object detection → VLM or VLM → gap → VLM) for more reliable outcomes.
Cost-efficient: Invoke the VLM only when warranted to reduce token usage and compute.
Fewer false alarms: Adds a time dimension (like dwell duration) and semantic checks before triggering events.
Flexible setup: Adjust thresholds, time-window sampling, and aggregation logic to fit different scenarios.
Great for: assessing disaster severity, checking whether an event affects specific conditions, or describing on-site situations.
✅ New Scenario Events
Person Dwelling : Detect and alert when a person remains within a defined zone beyond a configurable dwell time.
In/Out Count Alerts:
Trigger alerts based on net threshold breaches of person / vehicle / two-wheeler.
✅ VLM Playground
Blur Metrics
Add image-quality signals to prompts and evaluations: FFT-Blur and BRISQUE. Use these to reason about low-quality frames, or gate actions.
Multi-Image VLM Input
You’ll be able to submit multiple images (e.g., keyframes or an image set) to a single VLM query. This enables cross-frame consistency checks, better object re-identification, and richer scene understanding in one pass.
Build a VLM Detection
✅ Batch Edit Decoder & Compute Target
Batch edits can now specify decoder and compute targets, making it easier to change decoding and inference to the right resources for throughput targets.
✅ General Model Update — New Categories
Expanded class coverage to improve real-world object detections.
Vehicle Types Police Car, Tank Truck, Truck Crane, Container Truck, Flatbed Truck, Logistics Truck, Concrete Truck, Garbage Truck
Corrosion / Damage
Corrosion, Crack, Hole, Deformation
People Related
Male, Female, Person head Paramedic, Firefighter, Police officer, Worker
Infrastructure / Public Safety
Traffic light, Manhole, Garbage, Pothole
✅ Deep Link to Events (Event URL)
Opening a URL with an event log ID will directly pop the event details modal:
https://<your-domain>/observ/history?eventLogModalId={$EVENT_LOG_ID}
This is perfect for chat alerts, tickets, or cross-tool links into Observ.
✅ UI/UX Improvements
Create Task flow redesign for clearer steps and defaults.
Type-to-search added across more fields to speed up filtering and selection.
Coming Soon:
We’re preparing to roll out powerful new features to give you even more flexibility and control over task execution and resource usage:
Face & License Plate Blur (Privacy Anonymization) Automatic obfuscation for detected faces and license plates in events, configurable by task and retention policy.
Event Start/End Conditions Define separate start and end rules for each event (e.g., “starts when X happens for N seconds; ends when Y clears for M seconds”), enabling cleaner timelines, better deduping, and smarter notifications.
API Update (Versioned, Spec-Aligned) A refreshed, versioned API aligned to the latest requirements—including clearer schemas, consistent pagination & filtering, standardized error codes, and improved webhooks—so integrations are simpler and more reliable.
Stay tuned—these upgrades are just around the corner and will further enhance the scalability and efficiency of Observ in real-world deployments.
Support
For technical support, please contact our support team during business hours.
Last updated