---
license: apache-2.0
library_name: onnxruntime
pipeline_tag: image-segmentation
tags:
  - onnx
  - ultralytics
  - yolo11
  - instance-segmentation
  - speech-bubble
  - manga
---

# Manga Speech Bubble Segmentation — ONNX

ONNX export of [`huyvux3005/manga109-segmentation-bubble`](https://huggingface.co/huyvux3005/manga109-segmentation-bubble).

## Model

- Architecture: YOLO11n-seg
- Task: speech-bubble instance segmentation
- Classes: 1 (`speech-bubble`)
- Input: RGB tensor `[1, 3, 1600, 1600]`, float32, normalized to `[0, 1]`
- Outputs: detection/mask coefficients `[1, 37, 52500]` and mask prototypes `[1, 32, 400, 400]`
- ONNX opset: 17
- NMS: disabled; post-processing is required

## Export and validation

Exported with Ultralytics 8.4.118 on Apple Silicon using Python 3.12, then checked with ONNX Checker and executed with ONNX Runtime 1.28.0.

```python
from ultralytics import YOLO

model = YOLO("best.pt")
model.export(
    format="onnx",
    imgsz=1600,
    batch=1,
    opset=17,
    simplify=True,
    dynamic=False,
    nms=False,
)
```

## Attribution and data notice

The source model credits the MangaSegmentation and Manga109 datasets. This repository contains only the exported model and no dataset images or annotations. Please review and comply with the source model and dataset terms before using this model, especially for commercial distribution.

Source model: <https://huggingface.co/huyvux3005/manga109-segmentation-bubble>

## License

The source model card declares Apache-2.0. See the source repository and dataset terms for attribution and usage conditions.