Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Recognizing Artistic Style of Archaeological Image Fragments Using Deep Style Extrapolation

Gur Elkin, Ofir Itzhak Shahar, Yaniv Ohayon, Nadav Alali, Ohad Ben-Shahar

TL;DR

This work addresses recognizing artistic style from fragmented archaeological image fragments by introducing a two-module framework: a differentiable Style Extrapolation module that uses a modified autoencoder with a Gram-matrix based style loss and a masked content loss to fill fragment gaps, and a Style Classification module built on an EfficientNet backbone trained with cross-entropy. The training proceeds in two steps, first optimizing the extrapolator, then freezing it while training the classifier, enabling content-informed yet style-preserving representations. The approach achieves state-of-the-art results on CLEOPATRA and the novel POMPAAF dataset, with extensive analysis showing that fragment geometry and the number of pieces significantly influence performance, and that smaller, more numerous fragments can yield better accuracy due to fixed input resolution. The work contributes a new fragmentation-aware dataset, a novel two-part architecture, and practical insights for automated fragment reassembly in archaeology, offering a robust tool for reconstruction and content-based image retrieval in cultural heritage contexts.

Abstract

Ancient artworks obtained in archaeological excavations usually suffer from a certain degree of fragmentation and physical degradation. Often, fragments of multiple artifacts from different periods or artistic styles could be found on the same site. With each fragment containing only partial information about its source, and pieces from different objects being mixed, categorizing broken artifacts based on their visual cues could be a challenging task, even for professionals. As classification is a common function of many machine learning models, the power of modern architectures can be harnessed for efficient and accurate fragment classification. In this work, we present a generalized deep-learning framework for predicting the artistic style of image fragments, achieving state-of-the-art results for pieces with varying styles and geometries.

Recognizing Artistic Style of Archaeological Image Fragments Using Deep Style Extrapolation

TL;DR

This work addresses recognizing artistic style from fragmented archaeological image fragments by introducing a two-module framework: a differentiable Style Extrapolation module that uses a modified autoencoder with a Gram-matrix based style loss and a masked content loss to fill fragment gaps, and a Style Classification module built on an EfficientNet backbone trained with cross-entropy. The training proceeds in two steps, first optimizing the extrapolator, then freezing it while training the classifier, enabling content-informed yet style-preserving representations. The approach achieves state-of-the-art results on CLEOPATRA and the novel POMPAAF dataset, with extensive analysis showing that fragment geometry and the number of pieces significantly influence performance, and that smaller, more numerous fragments can yield better accuracy due to fixed input resolution. The work contributes a new fragmentation-aware dataset, a novel two-part architecture, and practical insights for automated fragment reassembly in archaeology, offering a robust tool for reconstruction and content-based image retrieval in cultural heritage contexts.

Abstract

Ancient artworks obtained in archaeological excavations usually suffer from a certain degree of fragmentation and physical degradation. Often, fragments of multiple artifacts from different periods or artistic styles could be found on the same site. With each fragment containing only partial information about its source, and pieces from different objects being mixed, categorizing broken artifacts based on their visual cues could be a challenging task, even for professionals. As classification is a common function of many machine learning models, the power of modern architectures can be harnessed for efficient and accurate fragment classification. In this work, we present a generalized deep-learning framework for predicting the artistic style of image fragments, achieving state-of-the-art results for pieces with varying styles and geometries.
Paper Structure (18 sections, 10 equations, 3 figures, 5 tables)

This paper contains 18 sections, 10 equations, 3 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Style Recognition of Fragments
  4. Style Transfer
  5. Data
  6. The Four Artistic Styles of Ancient Roman Wall Paintings
  7. POMPAAF Dataset Description
  8. Style Identification and Classification
  9. Style Extrapolation Module
  10. Style Classification Module
  11. Baselines
  12. Results
  13. Evaluation Metrics
  14. CLEOPATRA Challenge
  15. POMPAAF Dataset
  16. ...and 3 more sections

Figures (3)

  • Figure 1: Frescos from Pompeii, representing the four styles of ancient Roman wall paintings, sampled from the Pompeii Archives published by Napolitano et al. napolitano2023a.
  • Figure 2: Examples of the four fragmentation methods used to create POMPAAF. The methods are demonstrated on a fresco of the 4 Pompeian style, originally located in House of the Vettii, Pompeii. (a) Strictly square fragments. (b) Crossing cuts polygonal fragments. (c) Non-convex partition fragments. (d) Eroded Voronoi partition frgments.
  • Figure 3: Diagram of the proposed model. Yellow (top-left): Style extrapolation module. Green (top-right): Pre-trained image classifier. Purple (bottom): Loss computation for the style extrapolator. The depicted fragment is from the CLEOPATRA cascone2023classification training set.