LSI-108K

Interaction-derived supervision for spatial reasoning

LSI-108K is the dataset introduced in Spatial-Interactor: Learning Spatial Reasoning through Interaction with the Observable Physical World. It contains 107,518 verifiable QA pairs organized as a three-level spatial interaction curriculum.

Overview

Presentation

A three-level curriculum

Split	Samples	Curriculum role
L1	15,109	Passive world-state transitions
L2	69,487	Active self-state transitions
L3	22,922	Long-horizon transition integration
Total	107,518	Interaction-derived spatial QA

L1 follows object and robot state changes under a fixed observer. L2 follows camera motion and its spatial consequences. L3 integrates consecutive camera transitions into paths, turns, revisits, and reverse routes.

From trajectories to QA

Simulator actions, scene states, camera poses, object tracks, and robot trajectories provide geometric ground truth before task templates produce QA pairs. This keeps supervision tied to observable physical change.

Load

Annotations: all 107,518 QA pairs, organized into L1 / L2 / L3.

from datasets import load_dataset
qa = load_dataset("kagakouko/LSI-108K", "default")

Browse images: 66,153 QA pairs with original image bytes embedded. These are subsets of the full annotations, not additional training examples.

visual = load_dataset("kagakouko/LSI-108K", "images", split="train", streaming=True)
sample = next(iter(visual))
sample["images"][0].show()
print(sample["conversations"])

Image configuration	QA pairs
ai2thor	4,485
scene_images	4,233
bridgedata	10,000
arkitscenes	8,054
procthor	29,377
ai2thor_extra	10,004

hf download kagakouko/LSI-108K --repo-type dataset --include 'videos/*' --local-dir ./lsi108k
mkdir -p ./media
for archive in ./lsi108k/videos/*/*.tar; do tar -xf "$archive" -C ./media; done

The default configuration contains all annotations and relative media paths; images contains original image bytes. Questions, answers, and frame order are unchanged. The full release includes sources with restricted media access: 18,443 image QA and 13,833 video QA still require upstream media. See source access and licenses.

Data preparation and training.

The default configuration retains all annotations with portable source-relative media references. data/source_media_manifest.jsonl.gz maps each QA ID to upstream source identifiers without exposing local machine paths. Raw videos are not hosted here; obtain them from the original source under its terms.

Citation

For citation, use the project BibTeX.

License

The annotations are provided for research use. Media retain their source-specific licenses, including non-commercial restrictions where applicable. See source attribution and licenses.

data/source_media_manifest.jsonl.gz maps each QA ID to upstream source identifiers without exposing local machine paths. It does not grant media access; obtain restricted media from the upstream source under its terms.