Get Started

COCO8-Seg Dataset

Introduction#

Ultralytics COCO8-Seg is a small instance segmentation dataset composed of the first 8 images of the COCO train 2017 set, 4 for training and 4 for validation. This dataset is ideal for testing and debugging segmentation models, or for experimenting with new segmentation approaches. With 8 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.

Dataset Structure#

  • Images: 8 total (4 train / 4 val).
  • Classes: 80 COCO categories.
  • Labels: YOLO-format polygons stored under labels/{train,val} matching each image file.
  • Download size: ~1 MB.

Explore COCO8-Seg on Ultralytics Platform to browse every image with its polygon masks, view the class distribution and annotation heatmaps in the Charts tab, and clone it to train your own model in the cloud.

Dataset YAML#

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 COCO8-Seg dataset, the coco8-seg.yaml file is maintained at https://github.com/ultralytics/ultralytics/blob/main/ultralytics/cfg/datasets/coco8-seg.yaml.

ultralytics/cfg/datasets/coco8-seg.yaml
# Ultralytics 🚀 AGPL-3.0 License - https://ultralytics.com/license

# COCO8-seg dataset (first 8 images from COCO train2017) by Ultralytics
# Documentation: https://docs.ultralytics.com/datasets/segment/coco8-seg
# Example usage: yolo train data=coco8-seg.yaml
# parent
# ├── ultralytics
# └── datasets
#     └── coco8-seg ← downloads here (1 MB)

# Train/val/test sets as 1) dir: path/to/imgs, 2) file: path/to/imgs.txt, or 3) list: [path/to/imgs1, path/to/imgs2, ..]
path: coco8-seg # dataset root dir
train: images/train # train images (relative to 'path') 4 images
val: images/val # val images (relative to 'path') 4 images
test: # test images (optional)

# Classes
names:
  0: person
  1: bicycle
  2: car
  3: motorcycle
  4: airplane
  5: bus
  6: train
  7: truck
  8: boat
  9: traffic light
  10: fire hydrant
  11: stop sign
  12: parking meter
  13: bench
  14: bird
  15: cat
  16: dog
  17: horse
  18: sheep
  19: cow
  20: elephant
  21: bear
  22: zebra
  23: giraffe
  24: backpack
  25: umbrella
  26: handbag
  27: tie
  28: suitcase
  29: frisbee
  30: skis
  31: snowboard
  32: sports ball
  33: kite
  34: baseball bat
  35: baseball glove
  36: skateboard
  37: surfboard
  38: tennis racket
  39: bottle
  40: wine glass
  41: cup
  42: fork
  43: knife
  44: spoon
  45: bowl
  46: banana
  47: apple
  48: sandwich
  49: orange
  50: broccoli
  51: carrot
  52: hot dog
  53: pizza
  54: donut
  55: cake
  56: chair
  57: couch
  58: potted plant
  59: bed
  60: dining table
  61: toilet
  62: tv
  63: laptop
  64: mouse
  65: remote
  66: keyboard
  67: cell phone
  68: microwave
  69: oven
  70: toaster
  71: sink
  72: refrigerator
  73: book
  74: clock
  75: vase
  76: scissors
  77: teddy bear
  78: hair drier
  79: toothbrush

# Download script/URL (optional)
download: https://github.com/ultralytics/assets/releases/download/v0.0.0/coco8-seg.zip

Usage#

To train a YOLO26n-seg model on the COCO8-Seg dataset for 100 epochs with an image size of 640, you can use the following code snippets.

Train Example
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="coco8-seg.yaml", epochs=100, imgsz=640)

Sample Images and Annotations#

Here are some examples of images from the COCO8-Seg dataset, along with their corresponding annotations:

COCO8-seg instance segmentation dataset mosaic
  • Mosaiced Image: This image demonstrates a training batch composed of mosaiced dataset images. Mosaicing is a technique used during training that combines multiple images into a single image to increase the variety of objects and scenes within each training batch. This helps improve the model's ability to generalize to different object sizes, aspect ratios, and contexts.

Citations and Acknowledgments#

If you use the COCO dataset in your research or development work, please cite the following paper:

Citation
@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.

FAQ#

  • The COCO8-Seg dataset is a compact instance segmentation dataset by Ultralytics, consisting of the first 8 images from the COCO train 2017 set (4 for training, 4 for validation). This dataset is tailored for testing and debugging segmentation models or experimenting with new segmentation methods. It is particularly useful with Ultralytics YOLO26 for rapid iteration and pipeline error-checking before scaling to larger datasets.

  • To train a YOLO26n-seg model on the COCO8-Seg dataset for 100 epochs with an image size of 640, you can use Python or CLI commands. Here's a quick example:

    Train Example
    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="coco8-seg.yaml", 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, COCO8-Seg lets you run a training-and-validation pass to catch pipeline errors — a broken data loader, a misconfigured loss, or a bad augmentation — before committing to a larger dataset. Learn more about supported dataset formats in the Ultralytics segmentation dataset guide.

  • The YAML configuration file for the COCO8-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/coco8-seg.yaml. The YAML file includes essential information about dataset paths, classes, and configuration settings required for model training and validation.

  • COCO8-Seg (8 images) sits below COCO128-Seg (128 images) and the full COCO-Seg dataset (118,287 training images) in terms of size:

    • COCO8-Seg: 8 images (4 train, 4 val) — the fastest sanity check, ideal for CI and quick debugging.
    • COCO128-Seg: 128 images — balanced between size and diversity, with train and val sharing the same directory.
    • Full COCO-Seg: 118,287 training images — comprehensive but resource-intensive, requiring ~20 GB on first download.

    Use COCO8-Seg for the fastest possible pipeline check, then scale to COCO128-Seg or the full COCO-Seg dataset as confidence grows.

Comments6

Ask a question, share your results, or help someone else.

  1. kh-diabGitHub

    Hello I am struggling because I followed the same coco format in the yaml file but still get this error

    Epoch GPU_mem box_loss seg_loss cls_loss dfl_loss Instances Size

    0%| | 0/2 [00:00<?, ?it/s] Traceback (most recent call last): File "/home/kdiab/.conda/envs/yolov8/lib/python3.10/site-packages/ultralytics/utils/loss.py", line 289, in call targets = torch.cat((batch_idx, batch["cls"].view(-1, 1), batch["bboxes"]), 1) RuntimeError: Sizes of tensors must match except in dimension 1. Expected size 66 but got size 0 for tensor number 1 in the list.

    The above exception was the direct cause of the following exception:

    Traceback (most recent call last): File "/home/kdiab/Documents/OST/workspace/dataset/coding/Train_Model/train segment model V6.0.py", line 33, in results =model.train( File "/home/kdiab/.conda/envs/yolov8/lib/python3.10/site-packages/ultralytics/engine/model.py", line 652, in train self.trainer.train() File "/home/kdiab/.conda/envs/yolov8/lib/python3.10/site-packages/ultralytics/engine/trainer.py", line 204, in train self._do_train(world_size) File "/home/kdiab/.conda/envs/yolov8/lib/python3.10/site-packages/ultralytics/engine/trainer.py", line 381, in _do_train self.loss, self.loss_items = self.model(batch) File "/home/kdiab/.conda/envs/yolov8/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl return self._call_impl(*args, **kwargs) File "/home/kdiab/.conda/envs/yolov8/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl return forward_call(*args, **kwargs) File "/home/kdiab/.conda/envs/yolov8/lib/python3.10/site-packages/ultralytics/nn/tasks.py", line 101, in forward return self.loss(x, *args, **kwargs) File "/home/kdiab/.conda/envs/yolov8/lib/python3.10/site-packages/ultralytics/nn/tasks.py", line 283, in loss return self.criterion(preds, batch) File "/home/kdiab/.conda/envs/yolov8/lib/python3.10/site-packages/ultralytics/utils/loss.py", line 294, in call raise TypeError( TypeError: ERROR ❌ segment dataset incorrectly formatted or not a segment dataset. This error can occur when incorrectly training a 'segment' model on a 'detect' dataset, i.e. 'yolo train model=yolov8n-seg.pt data=coco8.yaml'. Verify your dataset is a correctly formatted 'segment' dataset using 'data=coco8-seg.yaml' as an example.

    and this is my yaml file

    Ultralytics YOLO , AGPL-3.0 license

    COCO8-seg dataset (first 8 images from COCO train2017) by Ultralytics

    Documentation: https://docs.ultralytics.com/datasets/segment/coco8-seg/

    Example usage: yolo train data=coco8-seg.yaml

    parent

    ├── ultralytics

    └── datasets

    └── coco8-seg ← downloads here (1 MB)

    Train/val/test sets as 1) dir: path/to/imgs, 2) file: path/to/imgs.txt, or 3) list: [path/to/imgs1, path/to/imgs2, ..]

    path: /home/kdiab/Documents/OST/workspace/dataset/model_train/Group-C-tiny-Segment # dataset root dir train: /home/kdiab/Documents/OST/workspace/dataset/model_train/Group-C-tiny-Segment/images/train # train images (relative to 'path') 4 images val: /home/kdiab/Documents/OST/workspace/dataset/model_train/Group-C-tiny-Segment/images/val # val images (relative to 'path') 4 images test: # test images (optional)

    Classes

    names: 0: weed 1: rumex

    and I am using model model = YOLO('yolov8x-seg.pt')

    my labeled files is as follows

    0 0.094970703125 0.46401098901098903 0.09033203125 0.160989010989011 0 0.07861328125 0.46153846153846156 0.0556640625 0.03296703296703297 0 0.116943359375 0.43763736263736264 0.02783203125 0.036813186813186814

    can you please advise what is the error

    1. Glenn JocherUltralytics

      @kh-diab it looks like the error you're encountering is due to a mismatch in the dataset format. Ensure that your dataset is correctly formatted for segmentation. Verify that your label files contain segmentation masks and not just bounding boxes. Also, make sure you are using the latest version of the Ultralytics package. If the issue persists, please check the dataset structure against the COCO8-Seg example provided in the documentation: https://docs.ultralytics.com/datasets/segment/coco8-seg/. This should help resolve the formatting issue.

  2. Mauricio566GitHub

    I really love their material. I have a question please help me. If the annotations of an image for a detection problem look like this... 1 0.5302913188934326 0.5031314492225647 0.2403690367937088 0.34829962253570557 So what does the annotation of a segmented image look like? because in the dataset I did not understand the annotation and you could help me understand this process. thank you

    1. Glenn JocherUltralytics

      @Mauricio566 hello! Thanks for your kind words and for reaching out with your question. 🌟

      For instance segmentation, unlike detection annotations which typically include class labels and bounding box coordinates (x_center, y_center, width, height), segmentation annotations are more complex. They include polygonal annotations or masks that precisely outline the object within an image.

      In datasets like COCO, each segmented object in an image will have a corresponding mask defined by a set of polygon points. These points outline the exact shape of the object rather than just a bounding rectangle. This allows for a more accurate and detailed representation, especially useful for applications requiring precise object localization and shape delineation.

      If you're working with the COCO8-Seg dataset, the annotations will include these detailed masks for each object in the images. You can view these annotations in the JSON format provided with the dataset, where each object's mask is represented as a series of coordinates outlining the object.

      I hope this helps clarify the difference! If you have any more questions or need further assistance, feel free to ask. Happy coding! 😊

      1
    2. jotar25GitHub

      this is actually not clear, should i separate masked color to multiple images? what is it the processs to generate labels from multi mask image?

    3. Glenn JocherUltralytics

      Hi @jotar25,

      Thank you for your question! 😊 Let's clarify the process for generating labels from multi-mask images for instance segmentation.

      In instance segmentation, each object in an image is annotated with a mask that precisely outlines its shape. These masks are typically represented as polygons or binary masks. Here's a step-by-step overview of the process:

      1. Separate Masks for Each Object: Each object in the image should have its own mask. If you have a multi-mask image where different objects are represented by different colors, you will need to separate these into individual masks.

      2. Polygon Annotations: For datasets like COCO, the masks are often stored as polygons. Each polygon is a list of coordinates that outline the object. These coordinates are stored in the dataset's annotation file (e.g., a JSON file for COCO).

      3. Binary Masks: Alternatively, masks can be stored as binary images where the pixels belonging to an object are set to 1, and all other pixels are set to 0. Each object will have a separate binary mask.

      4. Annotation Format: In the COCO format, each annotation includes:

        • segmentation: A list of polygons for the object.
        • bbox: The bounding box of the object.
        • category_id: The class of the object.
        • image_id: The ID of the image.

      Here's an example of how a COCO annotation might look:

      {
        "segmentation": [[x1, y1, x2, y2, ..., xn, yn]],
        "bbox": [x, y, width, height],
        "category_id": 1,
        "image_id": 123
      }
      Example Code for Generating Binary Masks

      If you have a multi-mask image and need to generate individual binary masks, you can use the following Python code snippet:

      import cv2
      import numpy as np
      
      # Load the multi-mask image
      multi_mask_image = cv2.imread('multi_mask_image.png')
      
      # Assume each object is represented by a unique color
      unique_colors = np.unique(multi_mask_image.reshape(-1, multi_mask_image.shape[2]), axis=0)
      
      # Create binary masks for each unique color
      binary_masks = []
      for color in unique_colors:
          mask = cv2.inRange(multi_mask_image, color, color)
          binary_masks.append(mask)
      
      # Save or process the binary masks
      for i, mask in enumerate(binary_masks):
          cv2.imwrite(f'binary_mask_{i}.png', mask)

      For more detailed information on the COCO8-Seg dataset and how to work with it, please refer to the COCO8-Seg Dataset Documentation.

      I hope this helps! If you have any more questions or need further assistance, feel free to ask. Happy coding! 🚀