YOLO26 Instance Segmentation • Virtual Calibrator 50.0mm

Automated Concrete Carbonation Depth & Irregularity Index

An advanced Deep Learning system designed to segment phenolphthalein-tested concrete prism samples, extract spatial scale factors, and compute continuous carbonation depth (dc) maps and invariant irregularity indices (IRIx).

Launch Analyzer Workspace Learn More
Concrete Carbonation Sample
Concrete Specimen Analysis Uncarbonated Alkaline Core Segmentation
Calibrated 50mm

1. Upload Image

Upload high-resolution photographs of phenolphthalein-stained concrete specimens. Supports drag-and-drop up to 32 MB.

JPG / PNG Auto-Resize

2. AI Segmentation

YOLO instance segmentation delineates specimen boundaries and alkaline cores with perspective distortion correction.

ONNX Engine Distance Map

3. Quantitative Metrics

Instantly extracts Irregularity Index (I), Core Area (mm²), and carbonation depth statistics (dc,mean, dc,max, dc,min, σdc).

CSV Export KPI Dashboard

ConcreteCARB Dataset Scale

Built upon manually annotated concrete carbonation specimens following standardized 72/18/10 train-valid-test distribution for robust evaluation.

0
Augmented Images
0 mm
Prism Width Reference
0 px
Native Resolution
0%
mAP50-95 Segmentation

Real-Time Carbonation Analyzer

Upload a concrete specimen photo. The ONNX engine will segment the specimen and uncarbonated core, computing continuous carbonation depth maps and irregularity metrics.

Drag and drop your image here

Or click to browse files from your computer (JPG, PNG)

Auto 50mm Calibration Perspective Correction

Don't have an image? Test with these samples:

Sample 1 SAMPLE 01
Sample 2 SAMPLE 02
Sample 3 SAMPLE 03
Sample 4 SAMPLE 04
Sample 5 SAMPLE 05
Sample 6 SAMPLE 06
Sample 7 SAMPLE 07
Sample 8 SAMPLE 08

About the Project

CarboDEPTH is an AI-powered platform designed for the automated quantification of carbonation depth and boundary irregularity in concrete prism specimens tested with phenolphthalein indicators.

Utilizing state-of-the-art YOLO26 Instance Segmentation combined with Euclidean Distance Transforms and automated spatial calibration (50.0 mm physical width), the system eliminates subjective manual scale measurements and provides reproducible statistics.

92.4%
mAP50-95 Seg
<30ms
ONNX Inference
50.0mm
Spatial Reference
Download Dataset & Paper (DOI) GitHub Repository

Privacy & Security Note: Uploaded images are stored in a temporary session directory and are automatically purged every 24 hours via a background scheduler.

Technical Specifications

  • Inference Engine: ONNX Runtime C++ Engine
  • Segmentation Model: YOLO26m Instance Segmentation
  • Spatial Calibration: Otsu adaptive thresholding + Minimum Area Bounding Rect (50.0 mm)
  • Perspective Mitigation: Aspect ratio evaluation (aspect ratio < 3.2) for side face compensation
  • Distance Transform: Exact L2 Euclidean Distance Map for continuous dc calculation

Research Team

Specialists and researchers behind dataset development, mathematical formulation, and model architecture.

Dr. José Alberto Guzmán Torres

Dr. José Alberto Guzmán Torres

Engineering Applications & AI

SIIIA MATH • UMSNH
jatechia.com.mx
Dr. Francisco Javier Domínguez Mota

Dr. Francisco Javier Domínguez Mota

Applied Math & Finite Difference

SIIIA MATH • UMSNH
Dra. Elia M. Alonso Guzmán

Dra. Elia M. Alonso Guzmán

Civil Engineering & Materials

UMSNH
Dr. Gerardo Tinoco Guerrero

Dr. Gerardo Tinoco Guerrero

Numerical Methods & Computation

SIIIA MATH • UMSNH
Dr. Wilfrido Martínez Molina

Dr. Wilfrido Martínez Molina

Civil Engineering & Materials

UMSNH

Graduate Researchers

Christopher Nolan M. B.

Christopher Nolan M. B.

Graduate Researcher

UMSNH
Eli C. I.

Eli C. I.

Graduate Researcher

UMSNH
Gabriela P. J.

Gabriela P. J.

Graduate Researcher

UMSNH
Jorge Luis G. F.

Jorge Luis G. F.

Graduate Researcher

UMSNH
Maybelin C. G. C.

Maybelin C. G. C.

Graduate Researcher

UMSNH