Segmentation of Maya hieroglyphs through fine-tuned foundation models

• DOI: 10.48550/arxiv.2405.16426
• Publication Date: 2024-05-26
• Authors: FNU Shivam; Megan Leight; Mary Kate Kelly; Claire Davis; Kelsey Clodfelter; Jacob Thrasher; Yenumula B. Reddy; Prashnna Kumar Gy...
• Keywords: Maya; Foundation (evidence)

The study of Maya hieroglyphic writing unlocks the rich history cultural and societal knowledge embedded within this ancient civilization's visual narrative. Artificial Intelligence (AI) offers a novel lens through which we can translate these inscriptions, with potential to allow non-specialists access reading texts aid in decipherment those hieroglyphs continue elude comprehensive interpretation. Toward this, leverage foundational model segment from an open-source digital library dedicated artifacts. Despite initial promise publicly available segmentation models, their effectiveness accurately segmenting was initially limited. Addressing challenge, our involved meticulous curation image label pairs assistance experts art history, enabling fine-tuning models. This process significantly enhanced performance, illustrating approaches value expanding dataset. We plan dataset for encouraging future research, eventually help make legible broader community, particularly heritage community members.