Toolkit Processor

Automatic tool detection using computer vision. Verify that every tool is in place with a single photo — powered by AI trained on your specific toolkit layouts.

Try it yourself — select a scenario:

Loading scenarios...

Get Started

Set up your own toolkits, train custom AI models, and deploy to iOS and macOS.

Free to try

How it works:

Define your toolkit layout
Place ArUco markers at corners
Take a photo to check tools
AI detects present & missing tools

Back to Sign In

All Toolkits

ID	Name	Template	Code	Status	Last Check-In	Actions

Create Template

Upload reference image

Upload an image of your toolkit to define tool positions

Template ID Lowercase, numbers, underscores

Template Name

Description

Fiducial Markers Type of markers used for perspective correction

Marker IDs (TL, TR, BR, BL)

Tools (0)

ID	Name	Template	Code	Location	Status	Tools	Actions

Current Status

Tool Status

Check-In History

Select Toolkit

Detection Mode ROI-based processes each predefined slot; Object Detection finds tools in the full image

Algorithm Select ROI detection algorithm

Detector Model Select trained object detection model (YOLO, etc.)

Upload Image

Click or drag image

Show detailed metrics

Detection Configuration

Configure detection mode, algorithm, and models for each template. These settings are also used by mobile and desktop apps.

Default Detection Mode & Algorithm

Configure which detection mode and algorithm to use by default for each template.

Loading templates...

ROI-based Detection (CNN Classifiers)

Active CNN model for ROI-based detection when CNN algorithm is selected.

Loading templates...

Object Detection (Detectors)

Active detector model for Object Detection mode.

Loading templates...

SSIM + Histogram Correlation

Compares current image to reference using Structural Similarity Index.

SSIM Present Threshold 0.45 SSIM score above this indicates tool is present

SSIM Missing Threshold 0.30 SSIM score below this indicates tool is missing

Histogram Correlation Boost 0.40 Good histogram correlation can compensate for borderline SSIM

Histogram Correlation Poor 0.10 Very low histogram correlation indicates mismatch

Normalized Diff Missing 0.28 Normalized pixel difference threshold for missing detection

Uniformity Analysis

Detects tools based on variance and texture in ROI (no reference needed).

Variance Empty Max 400 Max variance for empty slot

Variance Occupied Min 800 Min variance for occupied slot

Gradient Empty Max 15 Max gradient magnitude for empty

Gradient Occupied Min 30 Min gradient magnitude for occupied

Entropy Empty Max 4.0 Max color entropy for empty

Entropy Occupied Min 5.5 Min color entropy for occupied

Brightness Range Empty Max 80 Max brightness range for empty

Brightness Range Occupied Min 150 Min brightness range for occupied

Two-Reference Comparison

Compares current image to both "present" and "empty" references.

Uncertainty Threshold 0.01 SSIM difference below this is considered uncertain (tool status unclear)

API keys allow external applications (like the iOS app) to authenticate without user credentials. Keep these keys secure.

Template:

- Total Samples

- Verified

- Present

- Missing

Samples by Tool

Select a template to view tool statistics

Recent Samples

Select a template to view recent samples

Pending Verification

Review and verify training samples to improve data quality.

Select a template to view samples pending verification

Extract from Check-in History

Extract training samples from historical check-ins. Only high-confidence detections are included.

Toolkit:

Add Training Sample

Select Template

Select Tool

Tool State Label

Upload Image

Click to upload or drag image

Upload a cropped image of the tool slot

Samples added this session: 0 View All Stats →

Template:

Image:

Select a template, then upload a toolkit image

Click on ROIs to label them as present or missing

Tool ROIs

Select a template to see tools

Training Queue

- pending

Filter:

Status:

No queue items. Images captured from mobile devices will appear here.

Template:

- Total Models

- Active Model

- Training Samples

- Best Accuracy

Training Jobs

Background training jobs. Progress updates automatically.

Job ID	Architecture	Status	Progress	Metrics	Duration	Actions
No training jobs

CNN Classifiers

Binary classifiers that determine present/missing for each ROI slot. The active model is used for check-ins.

Version	Architecture	Accuracy	Precision	Recall	Size	Inference	Created	Actions
Select a template to view models

Detectors

Object detectors (YOLO, SSD, etc.) that locate and identify tools in a single pass. Best for varying tool positions.

Version	Architecture	mAP@0.5	mAP@0.5:0.95	Precision	Recall	Size	Inference	Created	Actions
Select a template to view models

Model Comparison

Compare model performance on test data.

Select a template and click "Compare All Models" to see comparison

Pending

Promoted

Rejected

False Positives

False Negatives

Pending Corrections

	Image	Tool	Original	Corrected	Algorithm	Date	Actions
No corrections found

Error Patterns

Problematic Tools

Tools with highest correction rates

Email	Name	Role	Status	Actions

Toolkit Processor

Toolkit Processor

Toolkit Processor

Try it yourself — select a scenario:

All Toolkits

Tools (0)

Add Tool

Calibrate Detection Thresholds

Samples Collected

Register New Toolkit

Current Status

Tool Status

Check-In History

Detection Configuration

Default Detection Mode & Algorithm

ROI-based Detection (CNN Classifiers)

Object Detection (Detectors)

SSIM + Histogram Correlation

Uniformity Analysis

Two-Reference Comparison

Create API Key

Samples by Tool

Recent Samples

Pending Verification

Extract from Check-in History

Available Check-ins

Add Training Sample

Tool ROIs

Quick Actions

Save Samples

Session Stats

Training Queue

Training Jobs

CNN Classifiers

Detectors

Model Comparison

Pending Corrections

Error Patterns

Problematic Tools

Pending Approvals 0

Add User

Edit Toolkit

Train New Model