Ra3i — Rock Art Analysis with Artificial Intelligence
Ra3i — Rock Art Analysis with Artificial Intelligence
AI-POWERED | TRACE. DETECT. CLASSIFY.

A photograph in.
A reviewable tracing out.
Seconds, not hours.

Speed up prehistoric rock art documentation with automated tracings, speedy motif classification and built-in editing tools made with archaeology specialists in mind. Analyze, archive and share your findings with ease — and shift precious time from digital processing back to FIELDWORK, CONSERVATION AND RESEARCH.

See how it works Request a demonstration

Already use Ra3i? Access it here.

{{ countSequences }}

distinct automatic processing sequences, growing as new techniques are added

{{ countSaved }}

average time saved from photograph to a usable first tracing

{{ countChecks }}

reviewable: each step, parameter and decision is documented as it happens, not written up afterwards

Rock art doesn't wait for us.

Fading
pigments
Unstable
rock faces
Human
threats

Panels fade under sunlight, flake from unstable rock faces, and are sometimes lost to vandalism or development before they are ever fully recorded. The single greatest constraint on documenting them has never been a shortage of will — it has been a shortage of time.

Ra3i is designed to adapt to the immense variability found in rock art across the globe, while keeping documentation consistent and searchable.

Tag filters
Substrate
Conservation Threats
Natural
Human
Equipment
Sites matching
Vandalism 11 sites
Urban Development 25 sites
Pick a tag to narrow the record down to the sites that need attention first.


That means collections as different as the trans-Saharan depictions of Sahel (Burkina Faso) and the complex narrative scenes of Serra da Capivara (Brazil) can be documented on a common, comparable basis.  Collaboration, cross-checking, and further research: Made easier.

Ra3i documents what can be seen in the rock: The panel, its motifs, and their characteristics. The interpretation of what those images mean remains with the archaeologist.

Drag the handle

There's almost nothing to see in this photograph. Pull the handle across and you get what Ra3i found in it: A faint painting persists.

From here it's the archaeologist's call: keep the lines that are right, fix the ones that aren't.

Faded rock art panel, photographed in the field
The same panel with the pigment Ra3i detected highlighted in red
The photo
What Ra3i found
Open air panel in Badajoz, Spain
Who Ra3i is for

Three contexts. One less thing to do by hand.

Ra3i removes the most time-consuming part of documentation so you can focus on what only your expertise can do.

01In the field
One panelField photograph
Painted panel photographed in the field

Archaeologists & rock art researchers

Document more before the site changes.

Tracing a panel can take hours at a desk. Ra3i turns the first pass into a reviewable result in seconds, giving researchers more time for fieldwork and interpretation.

First pass in under 30 seconds
Every step shown and explained
You review, interpret and decide
02Across a landscape
Site libraryMany panels
Recorded panel: red figures on yellow sandstone Recorded panel: red anthropomorphs on pale rock Recorded panel: white pigment panel Recorded panel: heavily weathered dark rock Recorded panel: painted bars in raking light Recorded panel: engraved figure on grey rock

Heritage & conservation authorities

Keep documentation consistent across sites.

Different surfaces, light and pigment make every photograph different. Ra3i adapts its processing to each case while keeping the workflow repeatable.

Learns from every case and improves
Consistent results across teams
Built for real-world conditions
03At scale
Classification overview12,842 images
Anthropomorph38%
Quadruped32%
Zoomorph18%
Other motifs7%
Indeterminate5%

Museums, universities & research institutions

Turn collections into comparable data.

Classification across thousands of images becomes practical. Results are cross-checked and reviewable, making large-scale research and pattern analysis easier than ever.

Scales to thousands of images
Crossed checked classifications
Export standardised, ready to use
How Ra3i works

A partner in rock art documentation.

Ra3i handles the grunt work so you can spend more time on what requires your expertise. See the what, why and how below, and document with confidence.

You
Your expertise guides every step.
Ra3i
Ra3i
Takes care of the mechanical work.
You
Review, interpret and decide.
Photograph
in
Reviewable
record out
Stage 01

Read the image

Understand the scene and surface

Before deciding how to process a photograph, Ra3i studies it: the quality of the light, the condition of the rock surface, the likely pigment, and roughly where the motifs sit on the panel. This builds a complete profile of the image that guides everything that follows.

What Ra3i sees
The panel as photographed Pigment activation map over the desaturated photograph Surface texture and erosion field Likely motif areas, bracketed
{{ visLabel }}
IMAGE CHARACTERISTICS
LightDAYLIGHT · UNIFORM
Pigmentred ochre · 0.81
TextureROUGH - WEATHERED
Composition7 CANDIDATE REGIONS
Stage 02

Build the tracing

Create a clean, accurate tracing

Ra3i chooses how to process the image automatically, drawing on a library of more than 40 domain-specialised filtering, masking, cleaning, and overlay techniques, and selecting the combination that fits this specific photograph. Every technique applied, parameter chosen, and intermediate result is recorded in a process log you can open, check, and edit.

Sequence applied · {{ carPos }}
The photograph as supplied The photograph after PCA decorrelation The raw pigment mask from k-means clustering The mask after speckle cleaning The finished tracing laid back over the photograph
{{ carLabel }}
* Final tracing manually corrected by an archaeologist.
Stage 03

Identify what's there

Detect and classify motifs

Ra3i identifies what's depicted — a zoomorph, an anthropomorph, and so on — by examining both the original photograph and the tracing it just produced. Because two independent analyses have to agree, every detection arrives with an explicit confidence score rather than a flat yes-or-no.

Detections on the tracing
The tracing Ra3i produced The same tracing with the detected motif regions boxed
{{ d.label }} {{ d.note }} {{ d.val }}
Your material stays yours. Images, tracings and process records produced with Ra3i are never published or shared — they remain under the control of the team or individual that produced them.
Memory

It gets better with use

When a new photograph resembles a case it has handled before, Ra3i can start from a previous solution rather than starting from scratch.

Case library
Run just finished
{{ memCtx }}
{{ memVerdict }}
{{ c.ctx }} {{ c.reuse }}
Runs Ra3i can't stand behind are discarded, not learned from.
Self evaluation

It checks its own work

Ra3i checks every step against the one before it, and rolls back anything that loses information.

Tracing vs. detected pigment{{ alignVal }}
The pigment activation map The tracing being matched onto the pigment map
Proposed cleaning{{ selfStatus }}
The traced mask before cleaning What the proposed cleaning would have left In red, the pigment that cleaning would have deleted The corrected mask after the step was reverted
{{ selfCap }}
X-AI

Nothing stays hidden

Every technique Ra3i applies is shown as an image explained in plain language: What it did, and why it was useful for this photograph. 

The photograph as supplied After PCA decorrelation The raw k-means pigment mask The mask after speckle cleaning The tracing laid back over the photograph The detected motif regions
{{ xaiStep }}
Applied

{{ s.title }}

Why

{{ s.why }}

Results & impact

Trace faster, trace more.

Ra3i cuts tracing time by 99.7%*, allowing teams to shift their attention from manual tracing to fieldwork, documentation, and the sites that need it most.

By applying domain-specific image analysis, Ra3i treats every image as a new challenge. It detects pigment, surface complexity, erosion, compositional details and over a dozen other indicators to build unique processing sequences, each tailored to best suit the panel in question.

Case in point: Serra da Capivara, Brazil
Thousands of years ago, these four zoomorphic figures were painted in the walls of Toca da Entrada do Baixão da Vaca, in Brazil.
To the right, see how Ra3i effortlessly detects pigment and produces a high-fidelity tracing in seconds.

* Based on an average of ~26 seconds per image under ideal conditions — balanced light, single pigment, a stable surface — measured against a 2 hours and 30 minutes baseline an experienced archaeologist typically spends tracing a comparable panel by hand. Harder photographs take longer, at both ends.

Panel Epsilon 6-4022 as photographed The same panel with the pigment Ra3i recovered highlighted The extracted tracing on its own The tracing with detected motifs boxed and labelled
{{ collageLabel }}
{{ zoomTitle }}

{{ zoomBody }}

Serra da Capivara panel — In order: photograph - detected pigment - extracted tracing - motif regions.
PROCESSING SEQUENCE · RED PIGMENT· 4 STEPS
Step 1 — contrast equalisation
01 Enhance
Step 2 — combined pigment mask
02 Mask
Step 3 — speckle removal
03 Clean
Step 4 — hole filling
04 Fill
Days
Hours
Minutes
Seconds

From panel to tracing

40+

domain-specialised image processing techniques, combined per photograph

Juggling exports
1 click

full-resolution exports of the tracing, overlays, and process record

What you get

Everything that comes back to the archaeologist.

Here’s what a Ra3i session delivers:

What the photograph is really like

Lighting, pigment, surface, and all other heuristics that decide how the image gets processed.

PNG
Visual analysis
images + analysis.json
IMG
Processing sequence
one or more images per step
Every step, on the record

The techniques and parameters Ra3i chose, open to inspection and to change.

The end views of a sequence

Transparent and ready to publish versions of the tracing, overlays and annotated figures.

PNG
Views
tracing · overlay · RGBA
JSON
Motif record
motifs.json
Motifs named and scored

Each detection, automatic or hand-written, as well as their characteristics.

Everything in one document

A single PDF that compiles the analysis, the sequence, the views and the motif record side by side.

PDF
Report
report.pdf
+
and more…
Intermediate results, noise layers and thumbnails
Beyond the single panel

Fuel data-driven research.

Every session a team runs also feeds a shared, searchable record of the art itself. Because motifs come back classified and scored the same way everywhere, panels stop being isolated files and start being comparable data.

On the map you can follow a specific motif types across a whole region, look for stylistic kinship between distant sites, or pull up panels recorded under comparable light, pigment, and surface conditions — questions that used to mean weeks in an archive.

Request a demonstration
Consortium partners

Built by software, archaeology, and heritage science together.

Lead partner
TechFrame

A Portuguese technology company that leads the consortium and is responsible for Ra3i's system architecture, software development, and the engineering behind every processing module.

Research partner
Instituto Politécnico de Tomar

Contributes the computational research behind image processing and motif classification, bridging archaeology, art history, and computer science.

Home, since December 2025, to Europe's first Chair of Rock Art Archaeology.

Scientific partner
Instituto Terra e Memória

Provides archaeological context, domain specific data, classification expertise, and scientific validation, ensuring every output meets the standards of the discipline it serves.

Founding partners

These archaeological teams and heritage bodies care for rock art across several countries. Whether they brought scientific rigor or supplied data, they have helped shaped RA3I into what it is today.

Fundação Côa Parque
Côa Parque Foundation
Portugal

Manages the Côa Valley Archaeological Park and Museum.

marioreis@arte-coa.pt
Instituto de Investigación Fundación Cueva de Nerja
Instituto de Investigación de la Cueva de Nerja
Spain

Research institute for the Nerja Cave, Andalusia.

luisefrenfr@gmail.com
FIRST-ART
FIRST-ART Project
Portugal / Spain

Multidisciplinary research on the Escoural and Maltravieso caves.

hipolitocollado@gmail.com
ArdalesTur
ARDALESTUR
Spain

Conservation and management of the Ardales Cave, Guadalteba.

info@ardalestur.es
Vila de Mação
Municipality of Mação
Portugal

Municipal archaeology supporting the Tagus Valley collections.

coimbra.rockart@yahoo.com
Université Joseph KI-ZERBO
Université Joseph KI-ZERBO
Burkina Faso

Department of History and Archaeology.

pon.coulibaly@ujkz.bf
Scotland's Rock Art Project
Scotland's Rock Art Project
Scotland

First major research project on Scottish prehistoric rock art.

jvaldeztullett@gmail.com
CAEAP
Archaeological Team CAEAP
Spain

Coordinates the Council of Europe's Cultural Route on Rock Art.

rmontes@prehistour.eu
+
You?
Anywhere

If you look after rock art, there's a place for you in the network.