docs/en/datasets/classify/imagenet.md
The Ultralytics ImageNet dataset (data="imagenet") is the ImageNet-1k / ILSVRC-2012 subset used to train and benchmark image classification models. It contains 1,000 object classes with 1,281,167 training images and 50,000 validation images at a 224x224 image size, and downloads to roughly 144 GB of data. The broader ImageNet database is far larger — over 14 million high-resolution images annotated with WordNet synsets across more than 20,000 categories — but Ultralytics trains on the standardized 1,000-class ILSVRC subset that became the de-facto benchmark for deep learning in computer vision.
{% include "macros/yolo-cls-perf.md" %}
imagenet dataset provides 1,000 classes with 1,281,167 training and 50,000 validation images (ILSVRC-2012), the standard pretraining benchmark for image classification.The Ultralytics ImageNet dataset uses the ILSVRC-2012 split:
| Split | Images | Classes |
|---|---|---|
| Train | 1,281,167 | 1,000 |
| Validation | 50,000 | 1,000 |
Images are stored in per-class folders named by WordNet synset ID (for example, n01440764), the layout Ultralytics classification training expects. Each of the 1,000 classes maps to a WordNet synset, and there is no separate test split, so the 50,000-image validation set is used to measure accuracy.
!!! note "Download size"
ImageNet-1k is a ~144 GB download, so make sure you have enough disk space before training. For quick experiments, the smaller [ImageNette](imagenette.md) and [ImageNet10](imagenet10.md) subsets use the same folder format and train in a fraction of the time.
The annual ImageNet Large Scale Visual Recognition Challenge (ILSVRC) let researchers benchmark algorithms on a large-scale, standardized dataset with consistent evaluation metrics. It drove major advances in deep learning for image classification, object detection, and other vision tasks — most notably AlexNet's 2012 win, which helped launch the modern deep-learning era.
The ImageNet dataset is widely used to train and evaluate deep learning models for image classification, object detection, and object localization. Landmark architectures such as AlexNet, VGG, and ResNet were all developed and benchmarked on ImageNet, and ImageNet-pretrained weights remain a common starting point for transfer learning across vision tasks.
To train a YOLO classification model on ImageNet for 100 epochs at a 224x224 image size, use the code snippets below. 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-cls.pt") # load a pretrained model (recommended for training)
# Train the model
results = model.train(data="imagenet", epochs=100, imgsz=224)
```
=== "CLI"
```bash
# Start training from a pretrained *.pt model
yolo classify train data=imagenet model=yolo26n-cls.pt epochs=100 imgsz=224
```
You can also manage classification datasets and run training in the cloud with Ultralytics Platform.
The ImageNet dataset spans the 1,000 ILSVRC-2012 classes, providing a diverse and extensive resource for training and evaluating computer vision models. Here are some example images from the dataset:
If you use the ImageNet dataset in your research or development work, please cite the following paper:
!!! quote ""
=== "BibTeX"
```bibtex
@article{ILSVRC15,
author = {Olga Russakovsky and Jia Deng and Hao Su and Jonathan Krause and Sanjeev Satheesh and Sean Ma and Zhiheng Huang and Andrej Karpathy and Aditya Khosla and Michael Bernstein and Alexander C. Berg and Li Fei-Fei},
title={ImageNet Large Scale Visual Recognition Challenge},
year={2015},
journal={International Journal of Computer Vision (IJCV)},
volume={115},
number={3},
pages={211-252}
}
```
We would like to acknowledge the ImageNet team, led by Olga Russakovsky, Jia Deng, and Li Fei-Fei, for creating and maintaining the ImageNet dataset as a valuable resource for the machine learning and computer vision research community. For more information about the ImageNet dataset and its creators, visit the ImageNet website.
The ImageNet dataset is a large-scale image database whose broader collection holds over 14 million high-resolution images annotated with WordNet synsets. In Ultralytics, data="imagenet" trains on the standardized 1,000-class ILSVRC-2012 subset, which is the de-facto benchmark for image classification pretraining. Landmark models such as AlexNet, VGG, and ResNet were trained and benchmarked on ImageNet, underscoring its role in advancing computer vision.
The Ultralytics imagenet dataset uses the ILSVRC-2012 subset with 1,000 classes, 1,281,167 training images, and 50,000 validation images at a 224x224 image size, for a total download of roughly 144 GB. The full ImageNet database is much larger (over 14 million images across more than 20,000 WordNet synsets), but the 1,000-class subset is the one used for classification training and benchmarking.
To train an Ultralytics YOLO model on ImageNet, load a pretrained classification model and point data at imagenet:
!!! example "Train Example"
=== "Python"
```python
from ultralytics import YOLO
# Load a model
model = YOLO("yolo26n-cls.pt") # load a pretrained model (recommended for training)
# Train the model
results = model.train(data="imagenet", epochs=100, imgsz=224)
```
=== "CLI"
```bash
# Start training from a pretrained *.pt model
yolo classify train data=imagenet model=yolo26n-cls.pt epochs=100 imgsz=224
```
For more in-depth training instruction, refer to our Training page.
Ultralytics YOLO26 pretrained models offer state-of-the-art performance in terms of speed and accuracy for various computer vision tasks. For example, the YOLO26n-cls model, with a top-1 accuracy of 71.4% and a top-5 accuracy of 90.1%, is optimized for real-time applications. Pretrained models reduce the computational resources required for training from scratch and accelerate development cycles. Learn more about the performance metrics of YOLO26 models in the ImageNet Pretrained Models section.
The annual ImageNet Large Scale Visual Recognition Challenge (ILSVRC) drove advances in computer vision by providing a competitive platform for evaluating algorithms on a large-scale, standardized dataset. Its consistent evaluation metrics fostered innovation in image classification, object detection, and image segmentation, continuously pushing the boundaries of deep learning and computer vision.