 |
|
|
|
| Vision and Learning Datasets |
Object/Scene Recognition
- Caltech-101:
101 categories, 30 training samples per category max.
- Caltech-256:
more of the same: 256 categories.
- NORB (large set):
NYU Object Recognition Benchmark: 5 categories, 10 instances per categories,
texture-less objects under many poses and illuminations. Highly variable
backgrounds, with small variation of scale and position (jittered-cluttered NORB).
- NORB (small set):
NYU rbject Recognition Benchmark: 5 categories, 10 instances per categories,
texture-less objects under many poses. 6 illuminations, uniform
backgrounds, no variation of scale and position (normalized-uniform NORB).
- Pascal Visual Object Classes Challenge datasets.
- LabelMe: segmented and labeled images from Antonio Torralba at MIT.
- Oxford Buildings dataset:
5062 images of 11 different Oxford landmarks.
- Image Parsing dataset
from Song-Chun Zhu's Lotus Hill Institute in China.
- Tiny Images:
1.5 million 32x32 images. The full set has 80 million images.
- ETH-80:
Objects from 8 categories, multiple instances, and multiple views on a blue background.
- Photo-tourism patches:
UBC/Microsoft aligned patches for invariant feature learning.
- Oxford Flowers.
- Priceton Event dataset: pictures of 8 sport
activities with about 200 images per category.
Face, People, and Car Datasets for Detection and Recognition
- faces: face-rec.org: huge list of links to Face recognition datasets and resources.
- faces: CMU Face datasets: PIE, face detection, facial expression.
- faces: AT&T/ORL
Face recognition dataset: 40 subjects, 10 images per subject (1992).
- faces: Purdue AR Face recognition dataset:
126 people, 4000 images, huge variations in appearance, lighting and expression.
- faces: Gallagher dataset from CMU:
can be used for detection and recognition.
- faces: Color FERET dataset from NIST:
8.5GB of faces. Can now be downloaded after an email request.
- faces: Yale Faces B:
10 subjects, 9 poses, 64 illuminations.
- faces: Essex face recognition dataset: Libor Spacek's dataset
of 395 subjects, 20 images per subject.
- people: INRIA Pedestrian Detection dataset (of HoG Dalal/Triggs fame).
- cars: UIUC Car detection dataset (from Dan Roth's group):
1050 training images (550 car and 500 non-car images); 170 single-scale test images, containing 200 cars at
roughly the same scale as in the training images; 108 multi-scale test images, containing 139 cars at various scales.
- cars: CMU Car detection dataset (from Henry Schneiderman):
104 images with 213 cars.
- cars, people, bikes: IG02, TU Graz dataset with INRIA
annotations: 365 bikes, 420 cars, 311 people.
- cars, people, bikes: TU Graz dataset: this dataset
was rendered obsolete by the INRIA dataset above.
Handwriting
- MNIST:
Handwritten digit dataset with 60,000 training samples and 10,000 test samples.
Video
Image retrieval
- Accio!:
content-based image retrieval dataset from WUSTL.
Pages with links to more datasets
|
|
|
|
|
|
