Researchers have introduced FlatLands, a new dataset and benchmark designed for completing bird's-eye view (BEV) floor maps from a single egocentric image. This dataset comprises over 270,000 observations from real indoor scenes, providing aligned data for observation, visibility, validity, and ground-truth BEV maps. The benchmark includes evaluation protocols for both in-distribution and out-of-distribution scenarios, testing various approaches including deterministic and stochastic generative models. FlatLands aims to serve as a rigorous testbed for uncertainty-aware indoor mapping and generative completion tasks relevant to embodied navigation. AI
IMPACT This dataset and benchmark could accelerate research in embodied navigation and generative AI for spatial understanding.
RANK_REASON The cluster describes a new dataset and benchmark for a computer vision task, presented in an academic paper. [lever_c_demoted from research: ic=1 ai=1.0]
- alphaXiv
- arXiv
- CatalyzeX
- Connected Papers
- CORE Recommender
- DagsHub
- FlatLands
- Gotit.pub
- Hugging Face
- Litmaps
- ScienceCast
- scite Smart Citations
- Subhransu S. Bhattacharjee
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
