---
license: cc-by-nc-sa-4.0
library_name: ultralytics
pipeline_tag: object-detection
datasets:
- M-Hamdy/SurgBox
tags:
- surgery
- medical
- cholec80
- cholecseg8k
- surgical-instruments
- yolo
extra_gated_prompt: >-
  These detectors were trained on Cholec80 and CholecSeg8k, which are distributed by their authors under their own
  terms. They may be used for non-commercial purposes only (CC BY-NC-SA 4.0), and their use is also subject to the
  Ultralytics license (AGPL-3.0).
extra_gated_fields:
  Name: text
  Affiliation: text
  I will use these models for non-commercial purposes only: checkbox
---

# SurgBox-Cross-Dataset

The three YOLO11-m detectors of the cross-dataset experiment in the SurgBox paper (Table 3). They detect the two instruments labelled in CholecSeg8k, Grasper and Hook, and are evaluated on the CholecSeg8k test videos (28, 43, 48, 52 and 55), which none of them saw in training.

**Paper:** *From Presence Labels to Bounding Boxes: Can Multimodal Large Language Models Scale Surgical Instrument Localization?* (MICAD 2026)<br>
**Code:** [github.com/M-Hamdy-M/SurgBox](https://github.com/M-Hamdy-M/SurgBox)<br>
**Dataset:** [M-Hamdy/SurgBox](https://huggingface.co/datasets/M-Hamdy/SurgBox)

## Checkpoints

COCO mAP on the CholecSeg8k test videos (2,560 frames, 3,228 boxes derived from the CholecSeg8k masks), paper Table 3.

| | File | Training labels | Frames | mAP | AP<sub>50</sub> | AP<sub>75</sub> | F1 |
|---|---|---|:---:|:---:|:---:|:---:|:---:|
| A | `yolo11m_cholecseg8k.pt` | CholecSeg8k masks (human) | 4,800 | 0.623 | 0.759 | 0.714 | 82.8 |
| B | `yolo11m_surgbox-2cls.pt` | SurgBox pseudo-labels | 79,564 | 0.833 | 0.973 | 0.929 | 95.1 |
| C | `yolo11m_surgbox-2cls_ft-cholecseg8k.pt` | SurgBox, fine-tuned on CholecSeg8k | 84,364 | 0.866 | 0.972 | 0.946 | 95.1 |

All three are YOLO11-m ([Ultralytics](https://github.com/ultralytics/ultralytics)), COCO-pretrained, trained at 640 px. C starts from B.

## Usage

```python
from huggingface_hub import hf_hub_download
from ultralytics import YOLO

model = YOLO(hf_hub_download("M-Hamdy/SurgBox-Cross-Dataset", "yolo11m_surgbox-2cls_ft-cholecseg8k.pt"))
results = model.predict("frame.png", imgsz=640, conf=0.25)
for box in results[0].boxes:
    print(model.names[int(box.cls)], box.xyxy[0].tolist(), float(box.conf))
```

Cholec80 and CholecSeg8k are not redistributed. Request Cholec80 from [CAMMA](http://camma.u-strasbg.fr/datasets); CholecSeg8k is available on [Kaggle](https://www.kaggle.com/datasets/newslab/cholecseg8k).

## License

Released for **non-commercial use only**, under [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/), consistent with Cholec80 and CholecSeg8k. The models were trained with Ultralytics, whose license ([AGPL-3.0](https://ultralytics.com/license)) also applies to their use.

## Citation

**If you use this model, please cite our paper:**

```bibtex
@inproceedings{hamdy2026surgbox,
  title     = {From Presence Labels to Bounding Boxes: Can Multimodal Large Language Models Scale Surgical Instrument Localization?},
  author    = {Hamdy, Mohamed and Abdel-Ghani, Muraam and Ahmed, Fatmaelzahraa and Nasar, Sifna and Ahmed, Mariam and Al-Jalham, Khalid and Al-Ali, Abdulaziz and Balakrishnan, Shidin},
  booktitle = {Medical Imaging and Computer-Aided Diagnosis (MICAD)},
  year      = {2026}
}
```

## Contact

Questions, issues and suggestions are welcome. Please open a discussion on this page or [contact me](mailto:mm1905748@qu.edu.qa).