datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
libero-3d-scene-pointcloud
LIBERO 3D Whole-Scene Point-Cloud GT (sim, labeled)
Per-frame whole-scene 3D point cloud for the LIBERO benchmark, sampled directly
from every object's visual mesh (posed by the replayed MuJoCo state) — complete,
view-independent geometry. Every point carries an instance label, and a single
is_ooi flag marks the task's objects-of-interest, so relevant objects are
extracted with one key at training time.
Scene = task objects + fixtures + the gripper (robot arm/mount excluded).… See the full description on the dataset page: https://huggingface.co/datasets/Kit-Key/libero-3d-scene-pointcloud.uavid-3d-scenes
UAVid-3D-Scenes
UAVid-3D-Scenes is a depth-estimation centric extension for the UAVid semantic dataset, organizing the original sequences based on the larger scenes they were captured in, providing undistorted RGB frames paired with metric depth maps obtained through COLMAP reconstructions and scaling.
📃 This dataset accompanies the paper TanDepth: Leveraging Global DEMs for Metric Monocular Depth Estimation in UAVs
License: CC BY-NC-SA 4.0 Creative Commons… See the full description on the dataset page: https://huggingface.co/datasets/hrflr/uavid-3d-scenes.3dscene_megadepth3dscene-arkitscene3dscene_kubric_movif3D_Room_Scene_Collections3D-SynthPlace_indoor_scenes_dataset
3D-SynthPlace indoor scene dataset in OptiScene (NeurIPS2025)
This is the 3D-SynthPlace dataset in the paper OptiScene: LLM-driven Indoor Scene Layout Generation via Scaled Human-aligned Data Synthesis and Multi-Stage Preference Optimization (NeurIPS2025).
3D-SynthPlace dataset JSON File Format Specification
The basic format of the scene description is JSON. This format is used to describe room floor and interior object layouts. You can refer prompts_all_scenes.json to… See the full description on the dataset page: https://huggingface.co/datasets/B3rrYang/3D-SynthPlace_indoor_scenes_dataset.3dscene-blendedmvs3dscene_dataset3dscene_spring3dscene_tartanair3dscene_wildrgbd3D-Front-Scene3dscene-arkitscene-highres3dscene-co3d288-Million-3D-Models-Scenes-Data-Sample
288-Million-Sets-3D-Models-Scenes-Data
Description
288 Million Sets - 3D Models & Scenes Data, which includes 270 million sets of 3D models and 18 million 3D scenes. 3D models are categorized into static models, interactive models and physics-enhanced models, covering various objects in indoor home environments such as furniture, appliances and kitchenware, etc. 3D scenes cover residential and commercial spaces, supports value-added services like component… See the full description on the dataset page: https://huggingface.co/datasets/Nexdata-AI/288-Million-3D-Models-Scenes-Data-Sample.3dscene_assetsc_cvit_val_3d_scene_control3D-Scene-Segmentation-HQ3D Segmentation HQ Dataset
The 3D Segmentation HQ dataset is a curated collection of 5 real-world scenes with high-quality object segmentation masks designed for research in 3D scene understanding, editing, and rendering.
This dataset improves upon existing benchmarks by providing cleaner and more consistent object masks across multiple views, enabling reliable evaluation and training for tasks such as:
3D semantic segmentation
Object-level scene editing (e.g., removal, recolorization)
3D… See the full description on the dataset page: https://huggingface.co/datasets/joshir/3D-Scene-Segmentation-HQ.3dscene__unreal4k3dscene-dl3dv3dscene-dynamic_replica3dscene_waymo3d-scene-poses3dscene-ase