---
license: apache-2.0
language:
  - en
tags:
  - motion-inbetweening
  - keypose-extraction
  - character-animation
  - adaptive-interpolation-synthesis
---

# Motion In-Betweening Dataset

## Dataset Description

This dataset contains character animation sequences from production. It is used for training and evaluating the motion in-betweening method presented in:

https://arxiv.org/abs/2605.02742

> Raël et al., "Adaptive Interpolation-Synthesis for Motion In-Betweening on Keyframe-Based Animation",
> SIGGRAPH 2026 Conference Papers (2026)

## Dataset Summary

The dataset provides character animation at 24 fps for a single production rig, with each frame represented as a D=596-dimensional vector of rig controller values. It is split into two subsets:

- **`in_house_dataset:`** 8.5 hours of internal production sequences, split 95% for training and 5% for held-out evaluation (`test_held_out_algorithmic` and `test_held_out_random` splits).
- **`prod_test_dataset:`** 5 minutes of production data with ground-truth blocking keyposes, used for the `test_production` evaluation split.

For each subset, the dataset provides the processed per-frame controller-value sequences (`vectorized_controller_values/`) along with the corresponding block keypose annotations (`vectorized_block_keyframes/`). For `in_house_dataset`, the block keyposes are determined by the Domain-Based Algorithm; for `prod_test_dataset`, they are ground-truth animator block keyposes.

## Usage

See the [GitHub repository](https://github.com/animaj-lab/mib-ais)

## Data Format

The repository is organized as follows:

```
root_directory/
├── in_house_dataset
│   ├── ranges.csv
│   ├── split_types.csv
│   ├── vectorized_block_keyframes
│   │   ├── test.h5
│   │   └── train.h5
│   └── vectorized_controller_values
│       └── ef25a8e5ecd2fa86420741e5646cd785aa025c44c06f4118aab77f558c8f6981
│           ├── test.h5
│           └── train.h5
└── prod_test_dataset
    ├── ranges.csv
    ├── split_types.csv
    ├── vectorized_block_keyframes
    │   ├── test.h5
    │   └── train.h5
    └── vectorized_controller_values
        └── ef25a8e5ecd2fa86420741e5646cd785aa025c44c06f4118aab77f558c8f6981
            ├── test.h5
            └── train.h5
```

## License

Apache 2.0