kingJulio's picture
fix task_categories warning (image-to-video)
a9374c6 verified
---
license: cc-by-4.0
pretty_name: Dream Engine example frames
size_categories:
- n<1K
task_categories:
- image-to-video
tags:
- synthetic
- test-fixture
- dream-engine
annotations_creators:
- machine-generated
---
# Dream Engine — example frames
Tiny synthetic test fixture used by the
[Dream Engine](https://dreamlabs.ai) Python SDK's ``dream.io`` module.
**This is not real teleop data** — start frames are colour gradients
and action sequences are uniform random in ``[-1, 1]``. The only
purpose is to give the bulk-inference docs + tests a real public
dataset to point ``frames_from_hf(...)`` at.
## Layout
| field | type | shape | content |
|---|---|---|---|
| ``row_id`` | string | — | ``synth_000`` – ``synth_015`` |
| ``start_frame`` | Image | 480 × 640 RGB | gradient |
| ``action_sequence`` | sequence-of-sequences[float32] | (48, 384) | random uniform |
## Used by
- The
[Bulk inference](https://dreamlabs.ai/docs/getting-started/bulk-inference)
quickstart in the Dream Engine docs.
- The
[Bulk augmentation example](https://dreamlabs.ai/docs/examples/bulk-augment-gr1)
walkthrough + companion Colab notebook.
- ``test_io_hf.py`` in the dream-engine SDK test suite.
## Replace this with your own data
Each row's ``start_frame`` should be a real RGB starting frame for the
spec you target (480 × 640 for ``dreamdojo-2b-gr1``); each
``action_sequence`` should be a real action sequence shaped
``(T, action_dim)`` matching the spec's expected control space.
Generated by ``scripts/seed_example_dataset.py`` in the
``dream-engine`` repo.