Crowdsource, Crawl, or Generate? Creating SEA-VL, a Multicultural Vision-Language Dataset for Southeast Asia
Abstract
Southeast Asia (SEA) is a region of extraordinary linguistic and cultural diversity, yet it remains significantly underrepresented in vision-language (VL) research. This often results in artificial intelligence (AI) models that fail to capture SEA cultural nuances. To fill this gap, we present SEA-VL, an open-source initiative dedicated to developing high-quality, culturally relevant data for SEA languages. By involving contributors from SEA countries, SEA-VL aims to ensure better cultural relevance and diversity, fostering greater inclusivity of underrepresented languages in VL research. Beyond crowdsourcing, our initiative goes one step further in the exploration of the automatic collection of culturally relevant images through crawling and image generation. First, we find that image crawling achieves approximately ~85% cultural relevance while being more cost- and time-efficient than crowdsourcing. Second, despite the substantial progress in generative vision models, synthetic images remain unreliable in accurately reflecting SEA cultures. The generated images often fail to reflect the nuanced traditions and cultural contexts of the region. Collectively, we gather 1.28M SEA culturally-relevant images, more than 50 times larger than other existing datasets. Through SEA-VL, we aim to bridge the representation gap in SEA, fostering the development of more inclusive AI systems that authentically represent diverse cultures across SEA.
Keywords
Cite
@article{arxiv.2503.07920,
title = {Crowdsource, Crawl, or Generate? Creating SEA-VL, a Multicultural Vision-Language Dataset for Southeast Asia},
author = {Samuel Cahyawijaya and Holy Lovenia and Joel Ruben Antony Moniz and Tack Hwa Wong and Mohammad Rifqi Farhansyah and Thant Thiri Maung and Frederikus Hudi and David Anugraha and Muhammad Ravi Shulthan Habibi and Muhammad Reza Qorib and Amit Agarwal and Joseph Marvin Imperial and Hitesh Laxmichand Patel and Vicky Feliren and Bahrul Ilmi Nasution and Manuel Antonio Rufino and Genta Indra Winata and Rian Adam Rajagede and Carlos Rafael Catalan and Mohamed Fazli Imam and Priyaranjan Pattnayak and Salsabila Zahirah Pranida and Kevin Pratama and Yeshil Bangera and Adisai Na-Thalang and Patricia Nicole Monderin and Yueqi Song and Christian Simon and Lynnette Hui Xian Ng and Richardy Lobo' Sapan and Taki Hasan Rafi and Bin Wang and Supryadi and Kanyakorn Veerakanjana and Piyalitt Ittichaiwong and Matthew Theodore Roque and Karissa Vincentio and Takdanai Kreangphet and Phakphum Artkaew and Kadek Hendrawan Palgunadi and Yanzhi Yu and Rochana Prih Hastuti and William Nixon and Mithil Bangera and Adrian Xuan Wei Lim and Aye Hninn Khine and Hanif Muhammad Zhafran and Teddy Ferdinan and Audra Aurora Izzani and Ayushman Singh and Evan and Jauza Akbar Krito and Michael Anugraha and Fenal Ashokbhai Ilasariya and Haochen Li and John Amadeo Daniswara and Filbert Aurelian Tjiaranata and Eryawan Presma Yulianrifat and Can Udomcharoenchaikit and Fadil Risdian Ansori and Mahardika Krisna Ihsani and Giang Nguyen and Anab Maulana Barik and Dan John Velasco and Rifo Ahmad Genadi and Saptarshi Saha and Chengwei Wei and Isaiah Flores and Kenneth Ko Han Chen and Anjela Gail Santos and Wan Shen Lim and Kaung Si Phyo and Tim Santos and Meisyarah Dwiastuti and Jiayun Luo and Jan Christian Blaise Cruz and Ming Shan Hee and Ikhlasul Akmal Hanif and M. Alif Al Hakim and Muhammad Rizky Sya'ban and Kun Kerdthaisong and Lester James V. Miranda and Fajri Koto and Tirana Noor Fatyanosa and Alham Fikri Aji and Jostin Jerico Rosal and Jun Kevin and Robert Wijaya and Onno P. Kampman and Ruochen Zhang and Börje F. Karlsson and Peerat Limkonchotiwat},
journal= {arXiv preprint arXiv:2503.07920},
year = {2025}
}
Comments
[SEA-VL Dataset] https://huggingface.co/collections/SEACrowd/sea-vl-multicultural-vl-dataset-for-southeast-asia-67cf223d0c341d4ba2b236e7 [Appendix J] https://github.com/SEACrowd/seacrowd.github.io/blob/master/docs/SEA_VL_Appendix_J.pdf