By default, it reuses the same 128 images for both training and validation to quickly demonstrate model "overfitting" as a proof of concept. Using COCO128 with YOLOv8/YOLOv11
Contains the first 128 images of the COCO train 2017 set.
Labels are provided in the YOLO TXT format , where each image has a corresponding .txt file containing normalized bounding box coordinates. download https //ultralytics.com/assets/coco128.zip
Includes all 80 object categories from the original COCO set, such as people, cars, animals, and common household items.
COCO128 is designed for speed without sacrificing class variety. By default, it reuses the same 128 images
For manual setups or offline use, you can download the zip directly: coco128.zip.
While users can manually download the asset, most modern Ultralytics workflows handle this automatically. Includes all 80 object categories from the original
The is a streamlined, 128-image subset of the massive Microsoft COCO (Common Objects in Context) collection. It is primarily used by developers and researchers as a "sanity check" to verify training pipelines and debug object detection models, such as Ultralytics YOLOv8 and YOLOv5, before committing to the full, resource-heavy 200,000+ image dataset. How to Download COCO128