docs/en/datasets/detect/objects365.md
The Objects365 dataset is a large-scale object detection benchmark with 1,742,289 training images and 80,000 validation images spanning 365 object classes — from people, cars, and chairs to bottles, dogs, and street lights. Created by Megvii researchers and presented at ICCV 2019, it focuses on diverse objects in the wild and is widely used to pretrain computer vision models that generalize better than ImageNet-pretrained ones.
<p align="center"> <iframe loading="lazy" width="720" height="405" src="https://www.youtube.com/embed/J-RH22rwx1A" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" allowfullscreen> </iframe><strong>Watch:</strong> How to Train Ultralytics YOLO on the Objects365 Dataset
</p>The Ultralytics Objects365.yaml configuration defines two splits:
| Split | Images | Description |
|---|---|---|
| Train | 1,742,289 | Labeled images for model training |
| Validation | 80,000 | Held-out images for evaluation and benchmarking |
The download retrieves the train and validation splits — 1,822,289 images in total — and the test: key in the configuration is left empty.
The Objects365 dataset supports a wide range of deep learning applications in object detection:
To label your own images, train, and manage large-scale datasets in your browser, run the full workflow with Ultralytics Platform.
The Objects365.yaml file defines the dataset configuration — the dataset paths, class names, and other metadata. It is maintained in the Ultralytics repository at https://github.com/ultralytics/ultralytics/blob/main/ultralytics/cfg/datasets/Objects365.yaml.
!!! example "ultralytics/cfg/datasets/Objects365.yaml"
```yaml
--8<-- "ultralytics/cfg/datasets/Objects365.yaml"
```
!!! warning "712 GB download"
Objects365 downloads automatically on first use and requires about 712 GB of free disk space — 345 GB of downloaded zip archives plus 367 GB for the extracted dataset. The download script installs the `faster-coco-eval` package and converts the annotations to YOLO format, which can take a long time depending on your connection and hardware.
To train a YOLO26n model on the Objects365 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.pt") # load a pretrained model (recommended for training)
# Train the model
results = model.train(data="Objects365.yaml", epochs=100, imgsz=640)
```
=== "CLI"
```bash
# Start training from a pretrained *.pt model
yolo detect train data=Objects365.yaml model=yolo26n.pt epochs=100 imgsz=640
```
The Objects365 dataset contains diverse, high-resolution images with dense bounding-box annotations across its 365 classes. The sample below shows the in-the-wild scenes and multi-object annotations typical of the dataset:
If you use the Objects365 dataset in your research or development work, please cite the following paper:
!!! quote ""
=== "BibTeX"
```bibtex
@InProceedings{Shao_2019_ICCV,
author = {Shao, Shuai and Li, Zeming and Zhang, Tianyuan and Peng, Chao and Yu, Gang and Zhang, Xiangyu and Li, Jing and Sun, Jian},
title = {Objects365: A Large-Scale, High-Quality Dataset for Object Detection},
booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
month = {October},
year = {2019}
}
```
We would like to acknowledge the team of researchers who created and maintain the Objects365 dataset as a valuable resource for the computer vision research community. For more information about the Objects365 dataset and its creators, visit the Objects365 dataset website.
The Objects365 dataset is used to train and evaluate object detection models in machine learning and computer vision. It provides 1,742,289 training images and 80,000 validation images across 365 object classes, and it is especially popular for pretraining detectors that are later fine-tuned on smaller, task-specific datasets.
The Ultralytics Objects365.yaml configuration covers 365 object classes split into 1,742,289 training images and 80,000 validation images — 1,822,289 in total, with no test split. The upstream release reports around 2 million images with 30 million bounding boxes overall.
Objects365 requires approximately 712 GB of disk space — about 345 GB of zip archives that download automatically the first time you train with data="Objects365.yaml", plus 367 GB for the extracted dataset. The download script installs the faster-coco-eval package and converts the annotations to YOLO format. You can browse smaller alternatives in the detection datasets overview.
To train a YOLO26n model using the Objects365 dataset for 100 epochs with an image size of 640, follow these instructions:
!!! example "Train Example"
=== "Python"
```python
from ultralytics import YOLO
# Load a model
model = YOLO("yolo26n.pt") # load a pretrained model (recommended for training)
# Train the model
results = model.train(data="Objects365.yaml", epochs=100, imgsz=640)
```
=== "CLI"
```bash
# Start training from a pretrained *.pt model
yolo detect train data=Objects365.yaml model=yolo26n.pt epochs=100 imgsz=640
```
Refer to the Training page for a comprehensive list of available arguments.
Objects365's 365-class vocabulary and dense annotations make it one of the strongest pretraining datasets for object detection — the ICCV 2019 paper reports a 5.6-point gain (42.0 vs 36.4 mAP) over ImageNet pretraining on COCO. Its images cover diverse real-world scenarios, which helps models generalize well to downstream detection tasks.
The YAML configuration file for the Objects365 dataset is available at Objects365.yaml. This file contains essential information such as dataset paths and class labels, crucial for setting up your training environment.