Datasets | Vision

Moorea Labeled Corals Related Project(s): Computer Vision Methods For Coral Reef Assessment The Moorea Labeled Corals dataset is a subset of the Moorea Coral Reef-Long Term Ecological Research (MCR-LTER) dataset packaged for Computer Vision research. It contains over 400.000 human expert annotations on 2055 coral reef survey images from the island of Moorea in French Polynesia. Each image has 200 expert random point annotations, indicating the substrate underneath each point.
Street View Text Related Project(s): GrOCR The Street View Text (SVT) contains 349 images harvested from Google Street View. Image text in this data exhibits high variability and often has low resolution. In dealing with outdoor street level imagery, we note two characteristics. (1) Image text often comes from business signage and (2) business names are easily available through geographic business searches. These factors make the SVT set uniquely suited for word spotting in the wild: given a street view image, the goal is to identify words from nearby businesses.
Yale Face Database Related Project(s): Doppelgänger List Comparison, Face Recognition The Yale Face Database (size 6.4MB) contains 165 grayscale images in GIF format of 15 individuals. There are 11 images per subject, one per different facial expression or configuration: center-light, w/glasses, happy, left-light, w/no glasses, normal, right-light, sad, sleepy, surprised, and wink.
Extended Yale Face Database B (B+) Related Project(s): Doppelgänger List Comparison, Face Recognition The Extended Yale Face Database B contains 16128 images of 28 human subjects under 9 poses and 64 illumination conditions. The data format of this database is the same as the Yale Face Database B.
Wet and Wrinkled Fingerprint Recognition Related Project(s): Wet and Wrinkled Fingerprint Recognition When fingers wrinkle in water, they become harder to recognize as similar to a dry finger, based on fingerprint scans. Vision lab has taken this problem forward by introducing the WWF(Wet and Wrinkled Finger) dataset and baseline performances on a NBIS matcher and a commercial algorithm. We have also proposed a classifier to distinguish wrinkled from non-wrinkled by using Wavelet features.
Volumetric Surface Texture Database Related Project(s): Texture Synthesis and Reconstruction Natural materials often exhibit complex reflectance and intricate geometry posing a real challenge in surface modeling. We have created the Volumetric Surface Texture Database to facilitate the study of these challenges. It contains a number of textures observed for varying view and illumination directions and was created at the Beckman Institute at the University of Illinois at Urbana-Champaign.
Coral Colony Segmentation And Area Measurement Tools Related Project(s): Computer Vision Methods For Coral Reef Assessment Size and growth of individual coral colonies underlies community population dynamics on the reef. However, non-invasively measuring three-dimensional surface areas of in-situ colonies is currently not practical. Two-dimensional planar area measurements from photographic projection can be a practical proxy for surface area, but this method often has significant methodological errors, reducing accuracy and replicability. Here, we present improvements in both hardware and software to improve accuracy for obtaining planar area from digital photographs. Using a monopod support with adjustable length reference positioning and a semi-automated image segmentation program, we achieved a coefficient of variation of 2.26% for repeated measurements of coral planar area in realistic ocean field conditions. This level of error is appropriate for rapid, inexpensive field studies of coral colony community size structure, tracking growth over time, and measuring the areal extent of coral bleaching or disease on individual colonies.
Honda UCSD Video Database Related Project(s): Face Recognition The goal of the Honda/UCSD Video Database is to provide a standard video database for evaluting face tracking/recognition algorithms. Each video sequence is recorded in an indoor environment at 15 frames per second, and each lasted for at least 15 seconds. The resolution of each video sequence is 640x480. Every individual is recorded in at least two video sequences. Since we believe that pose variation provides the greatest challenge to recognition, all the video sequences contain significant 2-D (in-plane) and 3-D (out-of-plane) head rotations. In each video, the person rotates and turns his/her head in his/her own preferred order and speed, and typically in about 15 seconds, the individual is able to provide a wide range of different poses. In addition, some of these sequences contain difficult events which a real-world tracker/recognizer would likely encounter, such as partial occlusion, face partly leaving the field of view, and large scale changes, etc.
Urban Tribes Related Project(s): Urban Tribes Image sharing via social networks has produced exciting opportunities for the computer vision community in areas including face, text, product and scene recognition. In this work we turn our attention to group photos of people at different social events. People can guess plenty of implicit information from the visual aspect of a group of people, but what can we automatically determine about the social subculture to which these people may belong? We propose a framework that integrates state-of-the art person and face detection and uses low- and mid-level features to capture the visual attributes distinctive to a variety of social groups. We proceed in a semi-supervised manner, employing a metric that allows us to extrapolate from a small number of pairwise image similarities to induce a set of groups that visually correspond to familiar urban tribes such as biker, hipster or goth. Automatic recognition of such information in group photos offers the potential to improve recommendation services, context sensitive advertising and other social analysis applications. Preliminary experimental results demonstrate our ability to categorize group photos in a socially meaningful manner.
Pacific Labeled Corals Related Project(s): Computer Vision Methods For Coral Reef Assessment Pacific Labeled Corals is an aggregate dataset containing 5090 coral reef survey images from four Pacific monitoring projects in Moorea (French Polynesia), the northern Line Islands, Nanwan Bay (Taiwan) and Heron Reef (Australia). Pacific Labeled Corals contain a total of 251,988 expert annotations across 4 pacific reef locations, and can be used as a benchmark dataset for evaluating object recognition methods and texture descriptors as well as for domain transfer learning research. The images have all been annotated using a random point annotation tool by a coral reef expert. In addition, 200 images from each location have been cross-annotatoed by 6 experts, for a total of 7 sets of annotations for each image. The paper detailing Pacific Labeled Corals is currently in submission. We will provide a link here upon publication.
YouTube Video Text Related Project(s): GrOCR YouTube Video Text (YVT) contains 30 videos. Each video has 15-second length, 30 frames per second, HD 720p quality and was collected from YouTube. The text content in the dataset can be divided into two categories, overlay text (e.g., captions, songs title, logos) and scene text (e.g. street signs, business signs, words on shirt).
UCSD Campus Images Dataset Related Project(s): Image Based Geolocalization Over 1400 images from the UCSD Campus containing 53 locations provided as a training set with 1204 images, a testing set with 272 images, and 66 images captured with Google Glass. All images are GPS tagged, except for the Google Glass images. All locations are provided with point coordinates.
Menu-Match Dataset Related Project(s): GrOCR The Menu-Match dataset include images of meals from three restaurants: an Asian restaurant offers a buffet-style setup where customers select 1-3 toppings that are served with a fixed serving size on top of brown or white rice; an Italian restaurant offers a variety of pizzas, lasagnas, and pastas, served with sides of breadsticks or salad; and a soup restaurant offers 10 soups with a side of one of 5 breads.