ArtInsight Dataset

ArtInsight Dataset: Precision Annotations for Easel Paintings' Deterioration Detection

Description of the dataset:
ArtInsight is a meticulously curated dataset tailored for the automated detection of deterioration in easel paintings, a central piece of our cultural and historical legacy. It comprises high-resolution full-frame images of these artworks, enhanced with detailed annotations provided by expert restorers that specifically highlight areas of wear and degradation. These precision annotations offer a fine-grained perspective for detection and analysis of the affected regions.

Designed with the computer vision community in mind, the ArtInsight dataset serves as a pivotal resource for deploying artificial intelligence techniques, especially deep learning, in the domain of art restoration. The robustness of the dataset has been tested using cutting-edge models such as Mask-RCNN and other pre-trained models, reinforcing its potential in automating and aiding the intricate tasks often reserved for human experts in art restoration.

By releasing ArtInsight, we aim to spearhead innovation in art conservation, restoration, and damage detection, offering the scientific community a reliable base for their research and development endeavors.
 

------------------------------------------------
DETAILS
------------------------------------------------

- We collected two types of "damages".

LPL: Lacune from the loss of the Painting Layer. “Being understood as a deterioration that causes a discontinuity across a surface” [25]. In our study case, there will be a loss where we can still see the support panel. 

Stucco: it is a mixture generally made with animal glue and calcium sulfate, which is used to fill lacunae caused by missing paint and to level these gaps with the paint surface, and then, the chromatic reintegration can be carried out on them 

Statistics:
- LPL: 8 images and 338 polygons
- Stucco: 12 images and 2571 polygons
* Total: 20 images and 2909 polygons

------------------------------------------------
LPL:
------------------------------------------------

- train: 6 images in JPG:
Pintura_0.3
Pintura_0.6
Pintura_0.12
Pintura_0.15
Pintura_0.18
Pintura_0.19
-*-*-*-*-*-*-
- And the corresponding JSON (lpl_train.json) with metadata (usefull for Machine Learning purposes).

- This JSON contains a total of 173 polygons labelled with tag "1"
-*-*-*-*-*-*-

- test: 2 images in JGP:
Pintura_0.10
Pintura_0.16
-*-*-*-*-*-*-
- And the corresponding JSON (lpl_test.json) with metadata (usefull for Machine Learning purposes).

- This JSON contains a total of 165 polygons labelled with tag "1"
-*-*-*-*-*-*-

------------------------------------------------
stucco:
------------------------------------------------

- train: 6 images in JPG:
Pintura_0.2
Pintura_0.4
Pintura_0.5
Pintura_0.7
Pintura_0.9
Pintura_0.13
Pintura_0.14
Pintura_0.17

-*-*-*-*-*-*-
- And the corresponding JSON (stucco_train.json) with metadata (usefull for Machine Learning purposes).

- This JSON contains a total of 1826 polygons labelled with tag "0"
-*-*-*-*-*-*-

- test: 2 images in JGP:
Pintura_0.1
Pintura_0.8
Pintura_0.11
Pintura_0.20

-*-*-*-*-*-*-
- And the corresponding JSON (stucco_test.json) with metadata (usefull for Machine Learning purposes).

- This JSON contains a total of 745 polygons labelled with tag "0"
-*-*-*-*-*-*-