README.md

metadata

pretty_name: SPARC VQA
tags:
  - robotics
  - vision-language
  - spatial-reasoning
  - sparc

SPARC VQA

SPARC VQA is the generated spatial VQA training dataset used in the SPARC Qwen3.5 model releases. Each example embeds its image bytes and includes a question, answer, task type, target type, source dataset identifier, split, and JSON metadata.

Raw unfiltered corpus: https://huggingface.co/datasets/irl-kit/SPARC-VQA-Raw

Ready-to-train split

Use train_filtered_t097_mpo700.parquet for SPARC-only training. This is the processed, release-ready dataset: it contains the exact 284,909 examples retained by the release configuration and needs no further SPARC filtering:

• selected_start_score >= 0.97
• sorted by selected_start_score in descending order before diversity selection
• a maximum of 700 (source, trajectory_name, subtask_index) groups per normalized object name
• vacant-goal and vacant-start examples mentioning gripper excluded

For the unfiltered 838,211-example source corpus, use irl-kit/SPARC-VQA-Raw. The raw repository contains train.parquet, the exact export_sparc_training_subset.py postprocessing script, release_mixture.yaml, and instructions for obtaining this processed dataset or creating a custom filtered variant.

Data schema

Field	Description
sample_id	Stable example identifier
image	Embedded image bytes and optional path
question	User text prompt
answer	Supervised assistant answer
target_type, task_type	Spatial target and VQA task labels
source, split, metadata	Provenance and generation metadata

Mixtures

Release	Data mixture
Qwen3.5-4B	SPARC VQA + FSD + RoboPoint + LLaVA-OneVision2
Qwen3.5-0.8B-VTFT	SPARC VQA + FSD + RoboPoint + LLaVA-OneVision2
Qwen3.5-9B-EO	SPARC VQA + FSD + RoboPoint + LLaVA-OneVision2 + EO-1.5M

FSD, RoboPoint, LLaVA-OneVision2, and EO-1.5M are external datasets and should be retrieved from their upstream releases under their respective terms. The released models are Qwen3.5-4B, Qwen3.5-0.8B-VTFT, and Qwen3.5-9B-EO.

Prompting compatibility

The released models are sensitive to output formatting. Use each model's bundled chat template with a single user turn containing the image followed by the question. For point prediction, use this suffix:

Output the point coordinates in JSON format like [{"point_2d": [x, y], "label": "target"}]. Use integer coordinates between 0 and 1000.

For trajectories or multiple points, use this suffix:

Return only a JSON list like [{"point_2d": [x1, y1], "label": "point_1"}, {"point_2d": [x2, y2], "label": "point_2"}, ...]. Use integer coordinates between 0 and 1000.

Citation

@article{blank2026sparc,
  title={SPARC: Reliable Spatial Annotations from Robot Demonstrations at Scale},
  author={Blank, Nils and others},
  journal={arXiv preprint arXiv:2606.13497},
  year={2026}
}