Researchers have introduced NavAble, a large-scale dataset and synthetic data generation pipeline designed to improve blind and low-vision users' navigation capabilities. The dataset includes over 41,000 curated images and 8,000 newly collected, annotated real-world images focusing on 11 accessibility-critical object classes. To further enhance scale and diversity, a synthetic data generation pipeline was developed, creating 452,000 frames across 37 environments. Experiments demonstrate that fine-tuning segmentation models on NavAble significantly improves object segmentation accuracy, with synthetic data augmentation yielding further gains. AI
IMPACT Enhances AI perception capabilities for assistive navigation technologies, potentially improving mobility for blind and low-vision individuals.
RANK_REASON The item describes a new dataset and synthetic data generation pipeline for a specific AI research task (blind navigation), published on arXiv. [lever_c_demoted from research: ic=1 ai=1.0]
- alphaXiv
- arXiv
- CatalyzeX
- DagsHub
- Gotit.pub
- Hugging Face
- NavAble
- SAM 3D
- ScienceCast
- vision-language model
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
