Spatial AI
spatialai-corpora
SpatialAi corpora
Training corpora for SpatialAi, a from-scratch model that understands and edits 3D scenes. Every record is a scene, a question or instruction, the tool calls that answer it, and the answer, labelled by an exact oracle.
Folder
Stage
Train
Val
Stress
S1_frames/
S1
20,000
2,000
2,000
S2_relations/
S2
20,000
2,000
2,000
S3_frames/
S3
20,000
2,000
2,000
S4_metric/
S4
20,000
2,000
2,000
S5_structure/
S5
20,000
2,000
2,000
S6_contact/
S6
20,000
2… See the full description on the dataset page: https://huggingface.co/datasets/khaines178/spatialai-corpora.GeoSR-Bench
GeoSR-Bench
Dataset and model weights for the paper:
Beyond Visual Fidelity: Benchmarking Super-Resolution Models for Large-Scale Remote Sensing Imagery via Downstream Task Integration [arXiv]
The code is available on GitHub: https://github.com/ai-spatial/GeoSR-Bench
Dataset Description
GeoSR-Bench directly connects super-resolution (SR) with downstream Earth monitoring tasks, moving beyond conventional fidelity-based evaluation. It comprises spatially co-located… See the full description on the dataset page: https://huggingface.co/datasets/ai-spatial/GeoSR-Bench.sea-small
Spatial Everyday Activities
[Website] [Contact]
Spatial Everyday Activities (SEA) is an egocentric dataset designed for training robotic foundation models. It comprises approximately 10,000 hours of egocentric data collected by computer vision experts across a diverse range of locations in the US and EU. SEA-small is a 100GB open-source subset of the full SEA dataset.
info@spatial-ai.com
Run the code
Setup an isolated environment
conda create -n sea python=3.12
conda… See the full description on the dataset page: https://huggingface.co/datasets/spatial-ai/sea-small.DERE
DERE Dataset
DERE is a multi-source ecosystem dataset for global carbon-flux prediction. It
integrates Ecosystem Demography (ED) simulations, ED-derived vegetation
structure, ESA CCI plant functional type fractions, LiDAR-derived forest-age
information, and real-world in-situ carbon-flux observations.
The dataset is organized into two complementary collections. GlobalMask
provides globally sampled simulation and remote-sensing data, while
InSituMatched links the same… See the full description on the dataset page: https://huggingface.co/datasets/ai-spatial/DERE.CarbonGlobe
CarbonGlobe: A Global-Scale, Multi-Decade Dataset and Benchmark for Carbon Forecasting in Forest Ecosystems
CarbonGlobe is a global-scale, multi-decade, machine-learning-ready dataset and benchmark for forecasting carbon dynamics in forest ecosystems. The dataset provides harmonized environmental drivers and carbon-related ecosystem outputs simulated by the Ecosystem Demography model version 3 (ED v3), enabling the development, evaluation, and comparison of deep learning models… See the full description on the dataset page: https://huggingface.co/datasets/ai-spatial/CarbonGlobe.libero_spatial_no_noops_lerobotv2.1