---
license: mit
pretty_name: SGE-Goal CALVIN-ABCD
task_categories:
- image-to-image
- robotics
language:
- en
size_categories:
- 10K<n<100K
tags:
- robotics
- manipulation
- goal-image-generation
- image-editing
- calvin
dataset_info:
  features:
  - name: id
    dtype: string
  - name: original_image
    dtype: image
  - name: edited_image
    dtype: image
  - name: initial_edit_region_mask
    dtype: image
  - name: final_edit_region_mask
    dtype: image
  - name: edit_instruction
    dtype: string
  splits:
  - name: train
    num_bytes: 2594140718.0
    num_examples: 22000
  - name: validation
    num_bytes: 114010800.0
    num_examples: 966
  - name: test
    num_bytes: 126468039.5
    num_examples: 1087
  download_size: 2828625558
  dataset_size: 2834619557.5
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
---

# SGE-Goal CALVIN-ABCD

Training and evaluation data for **SGE-Goal**, from *Goal Images Are Control Targets:
Spatially-Grounded Goal-Image Synthesis for Robotic Manipulation* (WACV 2027).

- Code: https://github.com/Riddhi-Chatterjee/SGE-Goal
- Project page: https://riddhi-chatterjee.github.io/SGE-Goal/
- Model checkpoints: https://huggingface.co/RiddhiCh/SGE-Goal

Each sample turns one atomic manipulation step from a robot episode into an image-editing example:
the scene before the step, the scene after it, where the edit happens, and the instruction. The
samples were extracted automatically from [CALVIN](https://github.com/mees/calvin), ABCD environments, without manual annotation.

## Fields

| Field | Type | Description |
|---|---|---|
| `id` | string | Sample identifier: the CALVIN episode and atomic segment, e.g. `episode_0_atomic_0001` |
| `original_image` | image | Observation before the manipulation step (`I_o`) |
| `edited_image` | image | Observation after the step, the goal image (`I_e`) |
| `initial_edit_region_mask` | image | Binary mask of the manipulated objects before the step (`M_i`) |
| `final_edit_region_mask` | image | Binary mask of where those objects are after the step (`M_f`) |
| `edit_instruction` | string | The atomic manipulation instruction (`T`) |

Images are 208 × 208 pixels; masks are white (edit region) on black.

The 1,087 test samples are the goal-generation inputs for the closed-loop CALVIN evaluation in the paper (GR-MG and GHIL-Glue); the code repository maps them back to CALVIN episodes.

## Splits

| Split | Samples |
|---|---|
| train | 22,000 |
| validation | 966 |
| test | 1,087 |

Samples are assigned to splits by MD5 hashing of their identifier (seed 42).

## Usage

```python
from datasets import load_dataset

ds = load_dataset("RiddhiCh/Calvin-Dataset-Final")
sample = ds["train"][0]
sample["original_image"]      # PIL.Image
sample["edit_instruction"]    # str
```

## How the data was extracted

The SGE-Goal data pipeline chains open vision models on each episode:

1. robot-arm masking and reference-frame selection (EVF-SAM, verified with BLIP-2),
2. class-agnostic object proposals on the reference frame (SAM),
3. object tracking through the episode (XMem),
4. separation of moved and stationary objects with LoFTR matching and 2-means clustering,
   followed by edit-mask generation and selection of the edited frame.

For episodes without a language instruction, VideoChat-Flash writes one from the video. The
pipeline is in [`data_pipeline/`](https://github.com/Riddhi-Chatterjee/SGE-Goal/tree/main/data_pipeline).
In a study with 50 annotators rating 500 tuples per dataset (1–5 scale), the lowest mean rating of
any extracted component across the three datasets was 4.44.

## Intended use

Training and evaluating goal-image generators and language-guided image editing for robotic
manipulation. Masks and edited frames come from an automatic pipeline: they are reliable in
aggregate, but individual samples can be imperfect, typically with heavily occluded targets or
several near-identical objects.

## Licence

CALVIN is released under the MIT License; this derived dataset is released under the same licence.

## Citation

If you use this dataset, please cite SGE-Goal and the source dataset. The SGE-Goal entry is
temporary until the proceedings are published.

```bibtex
@inproceedings{chatterjee2027goalimages,
  title     = {Goal Images Are Control Targets: Spatially-Grounded Goal-Image Synthesis for Robotic Manipulation},
  author    = {Chatterjee, Riddhi and Shrirao, Ritish and Mehrotra, Kushagra and Gopalakrishnan, Viswanath},
  booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
  year      = {2027}
}

@article{mees2022calvin,
  author  = {Oier Mees and Lukas Hermann and Erick Rosete-Beas and Wolfram Burgard},
  title   = {CALVIN: A Benchmark for Language-Conditioned Policy Learning for Long-Horizon Robot Manipulation Tasks},
  journal = {IEEE Robotics and Automation Letters (RA-L)},
  volume  = {7},
  number  = {3},
  pages   = {7327-7334},
  year    = {2022}
}
```