README.md

metadata

annotations_creators:
  - synthetic
language_creators:
  - no-annotation
language:
  - en
license:
  - other
multilinguality:
  - monolingual
pretty_name: Synthetic Reasoning Dataset
size_categories:
  - n<1K
source_datasets:
  - original
task_categories:
  - video-classification
task_ids:
  - multi-class-image-classification
paperswithcode_id: null

License

License: Other – Mixamo Restricted License (evaluation only, no ML training)

This dataset incorporates animations from Adobe Mixamo.
Mixamo’s license prohibits using its assets for training machine learning models.

Accordingly:

• ✅ You may use this dataset for evaluation, benchmarking, and non-commercial research.
• ❌ You may not use this dataset to train machine learning models or to create derivative datasets for training.
• All rights to the underlying animations remain with Adobe.

See the included LICENSE-Mixamo.txt for details.

---

Usage

This dataset is intended for benchmarking only.
It should not be used to train models due to Mixamo’s licensing restrictions.

Example use cases:

• Evaluating action recognition models (e.g., distinguishing Fighting vs. Dancing vs. Idle).
• Studying spatial reasoning in videos (Experiment 2).
• Testing models on trajectory understanding and viewpoint variation (Experiment 3).

Do not use this dataset as training data. If you require a trainable dataset, you must substitute animations that are licensed for ML use.

For testing with vision-LLMs see our GitHub repo.