README.md

---
configs:
- config_name: frame
  data_files:
  - split: train
    path: data/frame/train.parquet
  - split: test
    path: data/frame/test.parquet
- config_name: segment
  data_files:
  - split: train
    path: data/segment/train.parquet
  - split: test
    path: data/segment/test.parquet
- config_name: procedure
  data_files:
  - split: train
    path: data/procedure/train.parquet
  - split: test
    path: data/procedure/test.parquet
- config_name: all_tracks
  data_files:
  - split: train
    path: data/*/train.parquet
  - split: test
    path: data/*/test.parquet
license: other
license_name: orena-focus-data-usage-agreement
license_link: LICENSE
task_categories:
- visual-question-answering
language:
- en
tags:
- medical
- surgical
- video-understanding
- laparoscopy
- cholecystectomy
- foreign-objects
size_categories:
- 10K<n<100K
extra_gated_heading: "Request access to LapChole-FOCUS-VQA"
extra_gated_description: "Access is restricted to participants of the ORena FOCUS Challenge. Each request is reviewed manually; our team may take up to 3 days to process it. Requests from applicants who are not registered for the challenge, or who use a different email address than their challenge registration, will not be approved."
extra_gated_prompt: |
  Access to the dataset is currently only granted for participants of the ORena FOCUS Challenge (https://orena-focus-challenge.org). Further, data usage is only allowed under the condition that users agree to the following:
  (1) to use the data and videos provided only within the scope of the challenge and for the purposes of participating in the challenge,
  (2) to neither pass it on to a third party nor share it beyond members of the team,
  (3) to not publish the data or underlying annotations, or otherwise make them publicly available,
  (4) to refrain from any attempt to reidentify information that has been deidentified, such as individual surgeons or facilities, or the provenance of the video,
  (5) to maintain the data within a protected/secure environment compliant with HIPAA, GDPR, or similar regulation, and ensure access to the videos is restricted to members of the challenge team only.

  Algorithms and models generated by individual teams may be used for noncommercial or commercial purposes.
extra_gated_fields:
  I have registered for the ORena FOCUS Challenge using the same email address as for this Hugging Face account: checkbox
  I agree to the data usage agreement terms above: checkbox
extra_gated_button_content: "Request access"
---

<div align="center">

# LapChole-FOCUS-VQA

**A clinically grounded benchmark for long-context video understanding in minimally invasive surgery.**

[💻 Code](https://github.com/IMSY-DKFZ/orena-focus) &nbsp;•&nbsp; [🏆 Challenge](https://orena-focus-challenge.org/) &nbsp;•&nbsp; [⚖️ Data Usage Agreement](#license)

</div>

---

> [!IMPORTANT]
> ## 🔒 This is a gated dataset
>
> Access is granted **only to participants of the [ORena FOCUS Challenge](https://orena-focus-challenge.org)** and is subject to **manual review**. To be approved you must:
>
> 1. **Have a Hugging Face account** and be logged in — downloads are only enabled for registered, authenticated users.
> 2. **Be registered for the challenge** at [orena-focus-challenge.org](https://orena-focus-challenge.org).
> 3. **Use the same email address** for both your Hugging Face account and your challenge registration. Requests where the two do not match **cannot be approved**.
> 4. **Accept the data usage agreement** in the access request form.

---

## Overview

This is the second data batch in the [ORena FOCUS Challenge](orena-focus-challenge.org). It comprises:

 * 100 labeled laparoscopic cholecystectomies
 * 70 unlabeled laparoscopic cholecystectomies
 * 20,000 VQA pairs

It is fully compatible with the data format layed out in [the first batch](https://huggingface.co/datasets/orena-dkfz/heico-focus-vqa). 
For background on the clinical motivation and data generation please look into these resources.

---

## Evaluation Tracks

The benchmark uses a multi-track framework that systematically increases temporal and contextual demands:

| Track | Config name | Visual input | Description |
|-------|-------------|-------------|-------------|
| **Frame** | `frame` | Single frame | Tests short-context perception. No temporal modelling required. |
| **Segment** | `segment` | <= 5min clip | Tests understanding of motion and event context within a short window. |
| **Procedure** | `procedure` | Up to full video | Tests long-horizon reasoning over complete procedures lasting up to hours. |

---

## Dataset Structure

The dataset contains **100 labeled** and **70 unlabeled** laparoscopic cholecystectomy videos. QA annotations for the labeled videos are stored as parquet files, one per track and split:

```
data/
  frame/
    train.parquet
    test.parquet
  segment/
    train.parquet
    test.parquet
  procedure/
    train.parquet
    test.parquet
```

The `all_tracks` config merges the per-track files, making it easy to load the full benchmark in one call. Video files are hosted in the `videos/` folder of the repository.

---

## Dataset Schema

Each row in the dataset follows this schema:

| Field | Type | Description |
|-------|------|-------------|
| `id` | string | Unique question identifier |
| `video` | string | Video filename (e.g. `0015 - Heico - Rektum - 6.avi`) |
| `timestamp_start` | string | Start of the relevant time window (`HH:MM:SS`) |
| `timestamp_end` | string | End of the relevant time window (`HH:MM:SS`) |
| `procedure_type` | string | Surgical procedure name |
| `question` | string | The clinically grounded question |
| `answer` | string | Expert-validated ground-truth answer |
| `answer_format` | string | Expected answer format (e.g. `number`, `binary`, `time`, `fo_class`) |
| `primary_capability` | string | Primary capability (e.g. `object_identification`) |
| `secondary_capabilities` | list[string] | Additional capabilities required to answer correctly |
| `clinical_relevance` | bool | Whether the question is directly clinically relevant |
| `ood` | bool | Whether the question is considered out-of-distribution |

---

## Usage

> Access requires an approved gated request (see the box at the top). Once approved, authenticate with `huggingface-cli login` before loading.

### Using huggingface datasets

```python
from datasets import load_dataset

ds = load_dataset("orena-dkfz/lapchole-focus-vqa", "segment", split="test")
print(ds[0]["question"])   # "How many sponges are visible?"
print(ds[0]["answer"])     # "2"
```

### Using the orena-focus library

The [`orena-focus`](https://github.com/IMSY-DKFZ/orena-focus) library provides dataset loaders, answer-format parsing, and a full evaluation framework:

```commandline
pip install orena-focus
```

```python
from focus import FocusDataset, DatasetSplit, Track

ds = FocusDataset("lapchole", DatasetSplit.TEST, Track.SEGMENT)
request, reference = ds[0]
print(request.question)        # "How many sponges are visible?"
print(reference.answer)        # "2"
print(reference.format.type)   # "number"
```

To run inference with video input, download the video files first:

```python
from focus import FocusConfig, set_config, download

set_config(FocusConfig(root_dir="/data/focus"))
download("lapchole")  # downloads video files into /data/focus/lapchole/videos/
```

See the [library repository](https://github.com/IMSY-DKFZ/orena-focus) for end-to-end inference and evaluation examples.

---

## Citation

By the Challenge rules and the data usage agreement, you are not yet allowed to share this data beyond members of your team nor use it for any publication or for commercial uses. After the Challenge and the Challenge paper submission we will update the data usage agreement.

---

## Acknowledgements

LapChole-FOCUS-VQA was developed at the [Division of Intelligent Medical Systems (IMSY)](https://www.dkfz.de/en/imsy/), German Cancer Research Center (DKFZ), Heidelberg. The benchmark is the basis of the [ORena FOCUS challenge](https://orena-focus-challenge.org/) at MICCAI 2026. The project was partially funded through the [SAVE program](https://wellcomeleap.org/save/).

We gratefully acknowledge all annotation contributors, colleagues and domain experts whose effort made this benchmark possible.

---

## License

The dataset is not released under a standard open license. Use is governed by the **ORena FOCUS Data Usage Agreement**, which you must accept in the gated access request.