Object Detection Admin

-

Total Images

-

Labeled

-

Classes

-

Groups

System Health

Checking...

Dataset Overview

Loading...

Class Distribution

Quick Actions

Sample Data Populate all 8 detection scenarios with realistic demo data

Creates sessions, detections, vehicles, plates, crowd frames, infrastructure assets and animal observations — ready to explore in every admin section immediately.

Benchmark history

One number says whether generation is fast. A series says whether it is getting slower.

When	Seed	Sessions	Elapsed	Rows/s	Coverage
Not loaded.

Coverage Drift

Loading...

Scenario Quality

Loading...

📡 Live Detection Session

Detection Config

Session Name

Alert Threshold

–

FPS

–

Condition Score

–

Frames

–

Alerts

–

Objects Now

Active Sessions

Session ID	Status	Frames	Score	Alerts	Actions
No active sessions

📋 Detection Results

Time	Score	Priority	Objects	GPS	Conf
Loading…

🚨 Alerts

Time	Type	Severity	Score	Location	Ack
Loading…

Images –

Loading…

🖼

Select an image to start labeling

Click to segment · Shift+click = background point

Active Class

Click = foreground · Shift+click = background
Auto-segments on each click.

Mask cleanup:

Mask Preview

Score: —

Annotations (0)

Session Replay

Driver View

— — —

Frame Info

Select a frame on the slider

Confidence Calibration

Per-group confidence distributions from live detection. Accept the recommended threshold to update the group's detection sensitivity.

Click Refresh to load…

Alert Zones

Geographic polygons with custom alert thresholds. Detections inside a zone use the zone's threshold instead of the global one.

Name	Alert Threshold	Points	Actions
Loading…

Zone Map Click on the map to draw zone polygon points, then save

Detection Profiles

Named scenario presets that control which models and post-processors are active (color extraction, plate reading, vehicle classifier, crowd analytics). Attach a profile to a detection config to enable it.

Name	Scenario	Post-processors	Default	Actions
Loading…

Plate Statistics

—

Total Reads

—

Watchlist Hits

—

Unique Plates

Watchlist

Plate Reads

Watchlist only

Plate	Confidence	Session	GPS	Time	Match	Crop
Loading…

Vehicle Analytics

By Type

By Color

By Subtype

Crowd Analytics

—

Peak Persons

—

Avg Persons

—

Male %

—

Female %

Density Grid (latest)

Recent Vehicle Observations

Type	Subtype	Color	GPS	Confidence	Time
Loading…

—

Total Assets

—

Damaged / Faded

—

Total Observations

—

Asset Types

Assets by Type

Damaged / Faded Assets Condition score < 0.5

Loading…

Asset Inventory

Type	Label	OCR Text	Condition	Observations	GPS	First Seen
Loading…

—

Total Sightings

—

Road Hazards

—

Species

—

Days Tracked

Sightings by Species

Recent Road Hazards Heavy Alert Triggered

Loading…

Animal Observations

Hazards only

Species	Road Hazard	Confidence	GPS	Session	Time
Loading…

🎬 Detection Scenarios

Each scenario card represents a saved DetectionConfig linked to a DetectionProfile. Click Start to launch a live simulation from the saved config. Click Stop to end it. Use Create Scenario Configs to auto-generate all 8 scenarios (requires seeded profiles and sample data).

Loading scenarios… click Create Scenario Configs if none appear.

Label Review

Classes

Groups

Detection Classes

ID	Name	Labels in Dataset	Actions
Loading…

Class Groups / Scenarios

Groups bundle classes for specific detection scenarios. Each group gets its own trained model.
Raise per-class confidence to fix car↔truck or debris↔branches confusion.

Camera Library

Save RTSP streams, video files, or webcam URLs for use in detection configs

⚙ Detection Configs

Configs store all detection settings (cameras, model, active classes, thresholds). The default config is auto-loaded when starting a session.

Train New Model

Dataset Directory

Base Model

Epochs

Batch

Img Size

Device (empty=auto, 0=GPU)

Patience (early stop)

Resume from checkpoint

Mosaic, copy-paste, label smoothing and inverse-frequency class weighting applied automatically. Fixes car↔truck and debris↔branches confusion.

Every run records what it trained on — how many images, how many labelled, the class distribution and a hash of all three. Without it a run cannot be reproduced, because nothing says which data it saw.

Previous runs

Started	Status	Base model	Dataset	Hash
Not loaded.

📥 Use Detections for Training

Click ↻ to scan captured frames…

Auto-capture during detection

Automatically save every frame where objects are detected (no Start Capture needed). Applied to the next session you start.

Captured Frame Review

Dataset Statistics

Loading…

Training Progress

idle

Epoch 0 / 0 –

–

mAP@50

–

mAP@50-95

–

cls loss

–

seg loss

Training Log

Waiting for training…

📜 Training Run History

#	Group	Base Model	Epochs	Progress	mAP@50	Status	Best Model	Actions
Loading…

🤖 Trained Models

Select a model to assign it to a group. The group's detection config will auto-use that model.