docs/en/datasets/segment/coco128-seg.md
Ultralytics COCO128-Seg is a small but versatile instance segmentation dataset composed of the first 128 images of the COCO train 2017 set. This dataset is ideal for testing and debugging segmentation models, or for experimenting with new detection approaches. With 128 images, it is small enough to be easily manageable, yet diverse enough to test training pipelines for errors and act as a sanity check before training larger datasets.
labels/train2017 for the shared train and val image directory.!!! note
The default YAML points train and val at the same 128 images, so validation metrics measure fit on the training set rather than generalization on held-out data. Duplicate or customize the split if you need a true held-out set.
This dataset is intended for use with Ultralytics Platform and YOLO26.
A YAML file is used to define the dataset configuration. It contains information about the dataset's paths, classes, and other relevant information. In the case of the COCO128-Seg dataset, the coco128-seg.yaml file is maintained at https://github.com/ultralytics/ultralytics/blob/main/ultralytics/cfg/datasets/coco128-seg.yaml.
!!! example "ultralytics/cfg/datasets/coco128-seg.yaml"
```yaml
--8<-- "ultralytics/cfg/datasets/coco128-seg.yaml"
```
To train a YOLO26n-seg model on the COCO128-Seg dataset for 100 epochs with an image size of 640, you can use the following code snippets. For a comprehensive list of available arguments, refer to the model Training page.
!!! example "Train Example"
=== "Python"
```python
from ultralytics import YOLO
# Load a model
model = YOLO("yolo26n-seg.pt") # load a pretrained model (recommended for training)
# Train the model
results = model.train(data="coco128-seg.yaml", epochs=100, imgsz=640)
```
=== "CLI"
```bash
# Start training from a pretrained *.pt model
yolo segment train data=coco128-seg.yaml model=yolo26n-seg.pt epochs=100 imgsz=640
```
Here are some examples of images from the COCO128-Seg dataset, along with their corresponding annotations:
If you use the COCO dataset in your research or development work, please cite the following paper:
!!! quote ""
=== "BibTeX"
```bibtex
@misc{lin2015microsoft,
title={Microsoft COCO: Common Objects in Context},
author={Tsung-Yi Lin and Michael Maire and Serge Belongie and Lubomir Bourdev and Ross Girshick and James Hays and Pietro Perona and Deva Ramanan and C. Lawrence Zitnick and Piotr Dollár},
year={2015},
eprint={1405.0312},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
```
We would like to acknowledge the COCO Consortium for creating and maintaining this valuable resource for the computer vision community. For more information about the COCO dataset and its creators, visit the COCO dataset website.
The COCO128-Seg dataset is a compact instance segmentation dataset by Ultralytics, consisting of the first 128 images from the COCO train 2017 set. This dataset is tailored for testing and debugging segmentation models or experimenting with new detection methods. It is particularly useful with Ultralytics YOLO26 and Platform for rapid iteration and pipeline error-checking before scaling to larger datasets. For detailed usage, refer to the model Training page.
To train a YOLO26n-seg model on the COCO128-Seg dataset for 100 epochs with an image size of 640, you can use Python or CLI commands. Here's a quick example:
!!! example "Train Example"
=== "Python"
```python
from ultralytics import YOLO
# Load a model
model = YOLO("yolo26n-seg.pt") # Load a pretrained model (recommended for training)
# Train the model
results = model.train(data="coco128-seg.yaml", epochs=100, imgsz=640)
```
=== "CLI"
```bash
# Start training from a pretrained *.pt model
yolo segment train data=coco128-seg.yaml model=yolo26n-seg.pt epochs=100 imgsz=640
```
For a thorough explanation of available arguments and configuration options, you can check the Training documentation.
Because the download and train/val loop are much smaller than full COCO, COCO128-Seg lets you run a 1-epoch sanity check on a new pipeline — verifying the model trains, validates, and saves checkpoints correctly — before scaling to the full COCO-Seg dataset. Learn more about supported dataset formats in the Ultralytics segmentation dataset guide.
The YAML configuration file for the COCO128-Seg dataset is available in the Ultralytics repository. You can access the file directly at https://github.com/ultralytics/ultralytics/blob/main/ultralytics/cfg/datasets/coco128-seg.yaml. The YAML file includes essential information about dataset paths, classes, and configuration settings required for model training and validation.
COCO128-Seg (128 images) sits between COCO8-Seg (8 images) and the full COCO-Seg dataset (118,287 training images) in terms of size:
COCO128-Seg offers more diversity than COCO8-Seg while remaining far more manageable than the full COCO-Seg dataset for experimentation and initial model development.