JoyAI-Image-OpenSpatial: ~3M Visual-Spatial QA Samples

JoyAI-Image-OpenSpatial is a spatial understanding dataset built on the OpenSpatial framework. It contains approximately 3 million multi-turn visual-spatial question-answer samples aggregated from 7 open-source datasets and web data. The dataset was created by jdopensource and was last updated on April 15, 2026.

Use Cases

• Train models for 3D object grounding based on spatial understanding tasks.
• Develop systems for depth ordering reasoning based on multi-turn visual QA samples.
• Benchmark spatial relation reasoning capabilities based on the dataset's wide range of tasks.
• Improve distance estimation models based on the aggregated visual-spatial data.

Strengths

• Approximately 3 million multi-turn visual-spatial QA samples.
• Aggregated from 7 established open-source datasets: ARKitScenes, ScanNet, ScanNet++, HyperSim, Matterport3D, WildRGB-D, and Ego-Exo4D.
• Covers a wide range of spatial understanding capabilities as described.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is an approximation (~3M), which may limit precise suitability assessment.
• Specific file formats, data size, and license information are unknown.

Provenance

Source

jdopensource on Hugging Face.

Collection Method

Built on the OpenSpatial framework, aggregating data from 7 open-source datasets and web data.

Freshness

Last updated 2026-04-15 03:37:56; freshness should be verified.