images.cv Documentation
Everything you need to search, configure, and download free labeled image datasets.
Table of Contents
Introduction
images.cv is a free catalog of labeled image datasets for computer vision. Every dataset is free to download — no signup, no credit card, no usage limits. The download page lets you pick output size, color mode, train/val/test split, and augmentations before packaging a ZIP.
Search Datasets
Use the Search page to find datasets by keyword (e.g. "traffic lights", "cats"). Filter by minimum image count. You can also browse by category from the Categories index.
Configure the Download
- • Image size — 32, 64, 128, 224, 256, 512, or original ("nochange")
- • Color mode — RGB color or single-channel grayscale
- • Train / val / test split — sliders that sum to 100 (default 60 / 30 / 10)
- • Email me when ready — optional one-time download link over email
Augmentation
Pick any combination of 20 augmentations (rotate, flip, crop, perspective, blur, sharpen, brightness, contrast, saturation, color jitter, noise, cutout, JPEG compression, and more) and use the percentage slider to choose what fraction of the train split should be randomly augmented — from 0% to 100%. Augmented files are suffixed _aug so you can count them or filter them later.
Folder Structure
<dataset>/ images.cv_<dataset>/ data/ train/ <class_a>/*.jpg <class_b>/*.jpg val/ <class_a>/*.jpg <class_b>/*.jpg test/ <class_a>/*.jpg <class_b>/*.jpg meta.json
meta.json reference
Every ZIP ships with a meta.json file describing:
- • labels — the list of class names
- • amount_of_images — total image count
- • image_color_mode / image_size — what you picked
- • split_ratio + split counts — how images are distributed
- • augmentation — which augmentations ran and how many _aug files were produced
- • normalization — ImageNet mean/std to pair with transforms.Normalize
Loading in PyTorch / TF / Keras
from torchvision import datasets, transforms from torch.utils.data import DataLoader transform = transforms.Compose([ transforms.Resize((224, 224)), transforms.ToTensor(), transforms.Normalize( mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225], ), ]) train_ds = datasets.ImageFolder('./<dataset>/data/train', transform=transform) val_ds = datasets.ImageFolder('./<dataset>/data/val', transform=transform) train = DataLoader(train_ds, batch_size=32, shuffle=True) val = DataLoader(val_ds, batch_size=32)
For TensorFlow / Keras, use tf.keras.utils.image_dataset_from_directorypointed at the same data/train and data/val folders.
Custom B2B Datasets
Need production-grade datasets with custom classes, QA, annotation formats (YOLO, COCO, segmentation masks), or licensed commercial use? Our team builds them on demand — see the For Business page for scope, timeline, and pricing.
Support & FAQ
See the FAQ for quick answers, How to Use for a walkthrough, or Contact us directly — real humans answer.