Sparsh Garg Presents Mapillary Vistas Validation for Fine-Grained Traffic Signs at DataCV 2025
Our Sparsh Garg, a Senior Associate Researcher in the Media Analytics Department, will present “Mapillary Vistas Validation for Fine-Grained Traffic Signs: A Benchmark Revealing Vision-Language Model Limitations” at the Data Computer Vision (DataCV) 2025 workshop as part of ICCV 2025 in Honolulu, Hawai’i, on Sunday, October 19th, from 11:15 am – 11:25 am.
The paper, co-authored by Abhishek Aich, introduces MVV, a new dataset that provides fine-grained, expert-annotated labels for traffic signs, improving upon the coarse annotations in existing datasets like Mapillary. MVV includes pixel-level instance masks and distinguishes between semantically essential categories such as stop and speed limit signs. Benchmarking shows that DINOv2 consistently outperforms state-of-the-art vision-language models, establishing it as a strong baseline for fine-grained visual understanding in autonomous driving. DataCV is a workshop focused on advancing data-centric computer vision by exploring how datasets—rather than models—shape performance, fairness, and generalization in vision and vision-language systems.
