Download 74.2K Places labeled images
Choose output size, color mode, train/val/test split, and optional augmentations. We'll package your ZIP automatically.
74.2K images·ZIP download·No signup required·~38.0 MB
Image size
Choose the output size for all images.
Color mode
Choose the output color space.
Train / val / test split
Default is 70/15/15 (train / val / test). Move the sliders to pick your own.
Make your dataset more diverse by applying random transformations.
Note: ~30% of your train images will be augmented. For each one, one random augmentation is picked from your selection — so you can safely "check all" without stacking effects and ruining the image.
Example: Rotate
Hover an option below to preview its effect.
Download your dataset
Free to download. No signup required.
Download history & 1-click re-download
Save favorite datasets & augmentation presets
Faster support & updates
Cite this dataset (BibTeX)
@misc{imagescv_places_2026,
title = {Places Image Classification Dataset},
author = {{images.cv}},
year = {2026},
howpublished = {\url{https://images.cv/dataset/places-image-classification-dataset}},
note = {Labeled image dataset for computer vision research.}
}What's in your ZIP
images.cv_places/
├── meta.json # labels, split counts, augmentation info, ImageNet normalization
└── data/
├── train/ # ~70% of images
│ └── <class>/*.jpg
├── val/ # ~15% of images
│ └── <class>/*.jpg
└── test/ # ~15% of images
└── <class>/*.jpgQuick-start (PyTorch)
# Install
pip install torch torchvision
# Load the dataset (PyTorch ImageFolder)
from torchvision import datasets, transforms
from torch.utils.data import DataLoader
transform = transforms.Compose([
transforms.Resize((32, 32)),
transforms.ToTensor(),
])
train_ds = datasets.ImageFolder('./places/data/train', transform=transform)
val_ds = datasets.ImageFolder('./places/data/val', transform=transform)
train_loader = DataLoader(train_ds, batch_size=64, shuffle=True)
val_loader = DataLoader(val_ds, batch_size=64)
print(f'Classes: {train_ds.classes}')
print(f'Train: {len(train_ds)} images')
print(f'Val: {len(val_ds)} images')Suggested training settings
64
5e-5
30–60
Large dataset — you can train from scratch if you want, but transfer learning is still faster.
Did this dataset help your research?
images.cv is free forever. A small donation keeps the servers running.