Five real-world large-scale RGBD scenes with human-verified 3D object
annotations and spatial referring expressions, for evaluating
language-conditioned 3D object grounding in scenes far bigger than a
single room.
Released with FARM: Find Anything using Relational Spatial Memory —
paper ·
code ·
project page. The code repo's
EVALUATION.md documents how to reproduce the paper's grounding numbers on
this dataset out of the box (prebuilt scene graphs included below).… See the full description on the dataset page:
https://huggingface.co/datasets/GoldenGait/FARM-Scenes.
