Year: 2024
Type: preprint
Abstract: The rapid development of large Vision-Language Models (VLMs) has led to impressive results on academic benchmarks, primarily in widely spoken languages. However, significant gaps remain in the ability of current VLMs to handle low-resource languages and varied cultural contexts, largely due to a lack of high-quality, diverse, and safety-vetted data. Consequently, these models often struggle to understand low-resource languages and cultural nuances in a manner free from toxicity. To address these limitations, we introduce Maya, an open-source Multimodal Multilingual model. Our contributions are threefold: 1) a multilingual image-text pretraining dataset in eight languages, based on the LLaVA pretraining dataset; 2) a thorough analysis of toxicity within the LLaVA dataset, followed by the creation of a novel toxicity-free version across eight languages; and 3) a multilingual image-text model supporting these languages, enhancing cultural and linguistic comprehension in vision-language tasks. Code available at https://github.com/nahidalam/maya.
Source: arXiv (Cornell University)
Authors: Alam, Nahid, Kanjula, Karthik Reddy, Guthikonda, Surya, Chung, Timothy, Vegesna, Bala Krishna S, Das, Abhipsha, Susevski, Anthony, Chan, Ryan Sze-Yin, Uddin, S M Iftekhar, Islam, Shayekh Bin, Santhosh, Roshan, A, Snegha, Sharma, Drishti, Liu, Chen, Chaturvedi, Isha, Winata, Genta Indra, S, Ashvanth., Mukherjee, Snehanshu, Aji, Alham Fikri
Institutions: Cohere (Canada), Indiana University Bloomington, Imperial College London, Georgia Institute of Technology, Bangladesh University of Engineering and Technology, Indian Institute of Technology Bombay, University of Pennsylvania, Technische Universität Darmstadt, Indian Institute of Technology Dhanbad
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W4405255138