Dataset Hub

This dataset includes videos and images collected from YouTube and web crawlers, comprising fire and non-fire data. The non-fire video dataset includes chimney smoke, sunsets, and clouds. The fire video dataset includes flames, smoke, and fires. The fire image dataset includes fire and non-fire images.

USPS dataset, stored in HDF5 format, includes 7,291 training images and 2,007 test images, each 16 × 16 pixels, data type float32, labels int32.

This dataset was created by the team for the 2018 NASA Space Apps Challenge, aiming to develop a model capable of identifying images containing fires. The dataset includes two classes: fire images (755 outdoor fire images, some with dense smoke) and non‑fire images (244 natural images such as forests, trees, grasslands, rivers, etc.). The dataset exhibits class imbalance; it is recommended to keep equal numbers of each class in the validation set.

COYO‑700M is a massive dataset comprising 747 million image‑text pairs and various other metadata, intended for training diverse models. It is constructed by collecting alt‑text from HTML documents along with their associated images, aiming to support training of large foundation models and complement existing datasets.

This dataset is used for building detection and includes image‑related features such as image ID, the image itself, its width and height, and object information (area, bounding box, category, and ID). It consists of a training set with 5,000 samples. Download size is 547,367 bytes; total size is 1,427,880 bytes.

The dataset includes two primary features: `image` (image data) and `label` (classification label). The label covers four classes: CNV, DME, DRUSEN, and NORMAL. It consists of a training set with 480 samples. Total download size is 25,828,769 bytes; total size is 34,491,675 bytes.

The dataset consists of two primary features: images and labels. The image feature is of type image, and the label feature is of type categorical label, comprising seven categories: MEL, NV, BCC, AKIEC, BKL, DF, and VASC. The dataset is split into training, test, and validation subsets containing 10,015, 1,512, and 193 samples respectively. The download size is 3,244,182,740 bytes, and the total size is 3,203,813,771.0559998 bytes.

The UFPR‑VCR dataset was created by the Federal University of Paraná and contains 10,039 images covering 11 vehicle colors, primarily for vehicle color recognition research. Images are sourced from six publicly available license‑plate recognition datasets in Brazil, pre‑processed and filtered to ensure research‑suitable quality. The dataset emphasizes challenging scenarios such as nighttime and multiple viewpoints.

The Wild Bee Dataset was created by Berlin University of Applied Sciences and contains approximately 30 000 images of wild bees sourced from the iNaturalist database. It is primarily intended to support insect monitoring and species classification research. The dataset covers 25 common German wild bee species; four visually similar species were merged into a single class. During creation, the dataset underwent rigorous labeling, including segmentation masks for body parts. The goal is to assist biologists in annotating rare species using deep‑learning techniques, thereby improving understanding and protection of biodiversity.

This dataset consists of Sudoku images captured with smartphone cameras from various newspapers. It includes 200 Sudoku pictures, divided into a training set of 160 images and a test set of 40 images.

The MNIST Multiview Datasets comprise two four‑view datasets, each view being an R^14 × 14 vector. MNIST₁ generates four views by dividing the image into quadrants. MNIST₂ generates four overlapping views around the image centre, introducing redundancy between views.

This dataset is an image library in the old style of the manga Detective Conan, containing 117 characters and 27,104 images. Note that the image library may contain noise; preprocessing is recommended before use.

This dataset is primarily intended for image classification tasks, containing images and their corresponding class labels. The labels cover 553 distinct categories, each representing a specific organism or object.

Open Images is a dataset containing approximately 9 million image URLs, annotated with over 6 000 category labels. The dataset is split into training and validation sets, and each image may have one or multiple labels; label information can be obtained via CSV files.

Open Images is a dataset containing approximately 9 million images annotated with over 6,000 category labels. The dataset is provided by Google under a CC BY 4.0 license and is split into training and validation sets, each image having a unique 64‑bit ID and possibly multiple labels.

The dataset comprises approximately 29,843 cat‑face images of size 64 × 64 pixels, sourced from multiple open datasets and preprocessed to ensure that each image contains only a cat face.

The Motorbikes (Side) dataset was collected by undergraduate students at Caltech in February 2001 from the web. It contains 826 side‑view images of motorcycles in JPEG format. Additionally, the dataset includes a MATLAB file ImageData.mat, which contains an 8 × 826 matrix SubDir_Data; each column stores the coordinates of a motorcycle in the image in the format [x_bot_left y_bot_left x_top_left y_top_left x_top_right y_top_right x_bot_right y_bot_right].

This project stores the dataset used for a flower‑recognition project based on a convolutional neural network. The original dataset is publicly available on Kaggle but is inconvenient to download; it is provided here for easy access. The data also serve as backend content for a web‑frontend that displays scrolling images.

The Stanford Dog Dataset contains approximately 20,000 images spanning 120 categories, with each image accompanied by corresponding annotations. The dataset is used to train convolutional neural network (CNN) classifiers; due to the limited data volume, transfer learning techniques are employed, utilizing pretrained models such as VGG16.

In this work, we construct a large‑scale logo dataset, Logo‑2K+, which encompasses a variety of logo categories extracted from real‑world logo images. Our logo dataset contains 167,140 images, divided into 10 root categories and 2,341 sub‑categories.

This dataset contains 764 images for detecting two classes: wearing helmets and not wearing helmets. The images are annotated with bounding boxes in PASCAL VOC format.

The dataset contains 5,376 annotated images covering seven categories of urinary sediment particles: cast, cryst (crystals), epith (epithelial cells), epithn (epithelial nuclei), eryth (erythrocytes), leuko (leukocytes), mycete.

UJM TIV is a material dataset that contains images of 11 material categories, each image being a 20 × 20 RGB picture. The dataset aims to provide images with higher intra‑class variability, captured from four different viewpoints and four lighting conditions.

This repository contains a Python script for generating a 2D geometric shapes dataset, along with the dataset itself. The dataset includes 16 different geometric shapes, each randomly oriented and positioned within 224 × 224 pixel images.

This is an image dataset for wildlife target detection in mixed conifer‑broadleaf forests. It contains 25,657 images captured by infrared cameras in the Northeast Tiger and Leopard National Park, covering 17 major species (15 wildlife and 2 livestock): Northeast tiger, Northeast leopard, wild boar, sika deer, spotted deer, Asian black bear, red fox, Asian badger, raccoon dog, musk deer, Siberian weasel, ferret, yellow‑throated marten, leopard cat, Manchurian hare, cattle, and dog. All images are annotated in Pascal VOC format with resolutions of 1280 × 720 or 1600 × 1200 pixels.

A newly collected large‑scale fake‑image dataset for evaluating human and model ability to distinguish AI‑generated visual content.

The dataset is primarily used for image classification tasks and contains 27 different painting‑style labels. It includes 6,417 images split into training, validation, and test sets. Preprocessing automatically orients images and resizes them to 416 × 416 pixels.

GRIT is a large‑scale image‑text pair dataset built on COYO‑700M and LAION‑2B, focusing on precise grounding of text to image regions. The dataset extracts and links textual fragments (such as noun phrases and referring expressions) to corresponding image areas, supporting tasks including image captioning, visual question answering, object detection, and zero‑shot classification. Each data instance includes detailed image and text information, along with metadata such as image dimensions, text description, and similarity scores between text and image.

The dataset contains two main features: images and labels. Labels cover 16 different clothing categories, including Blazer, Blouse, Dress, etc. It is divided into a training set (4,283 samples) and a test set (756 samples). The download size is 139,594,572 bytes, and the total size is 138,280,643.011 bytes.

Imagewoof is a subset of ImageNet containing ten dog‑breed categories, designed to provide a challenging image‑classification benchmark. Created by Jeremy Howard, it is released under the Apache‑2.0 license. The dataset includes images and corresponding labels, with a defined train/validation split.

# Dataset Card for Naruto BLIP captions _Dataset used to train [TBD](TBD)._ The original images were obtained from [narutopedia.com](https://naruto.fandom.com/wiki/Narutopedia) and captioned with the [pre-trained BLIP model](https://github.com/salesforce/BLIP). For each row the dataset contains `image` and `text` keys. `image` is a varying size PIL jpeg, and `text` is the accompanying text caption. Only a train split is provided. ## Example stable diffusion outputs  > "Bill Gates with a hoodie", "John Oliver with Naruto style", "Hello Kitty with Naruto style", "Lebron James with a hat", "Mickael Jackson as a ninja", "Banksy Street art of ninja" ## Citation If you use this dataset, please cite it as: ``` @misc{cervenka2022naruto2, author = {Cervenka, Eole}, title = {Naruto BLIP captions}, year={2022}, howpublished= {\url{https://huggingface.co/datasets/lambdalabs/naruto-blip-captions/}} } ```