Computer vision datasets [closed]

Closed. This question does not meet Stack Overflow guidelines. It is not currently accepting answers.

---

Questions asking us to recommend or find a tool, library or favorite off-site resource are off-topic for Stack Overflow as they tend to attract opinionated answers and spam. Instead, describe the problem and what has been done so far to solve it.

Closed 9 years ago.

Improve this question

I am testing computer vision algorithms for image categorization. I would like to find a dataset with a few categories of objects e.g. cats and dogs. This dataset should have all the variability within each class be due to the class's intrinsic variability. That is, I don't want to have to worry about pictures taken from different viewpoints or under different lighting conditions etc. Almost all the variability within a category should be due to the intrinsic variability of that category e.g. the category of cats would contain many different images because cats actually look different from one another, not because the images we开发者_C百科re produced under different conditions.

Preferably, the objects will be "cut out" (on a uniform background). The size of the dataset is not important. Synthetic images (perhaps produced with 3D graphics software) are also ok. The images need to come labeled with their category.

Does anyone know of a dataset like this?

---

Computer Vision Online seems to have a very comprehensive list of CV and image processing datasets: http://computervisiononline.com/datasets

---

I tried SVMs a few years ago with this library:

http://www.cs.columbia.edu/CAVE/software/softlib/coil-100.php

---

Try the caltech 256 dataset: http://www.vision.caltech.edu/Image_Datasets/Caltech256/

It will give you a good basis of comparison against other algorithms used in the community as well.

---

I've listed some of the image data sets that we've been working on during my PhD time - but you should find really plenty of them on the web. From what you described you are looking for an object recognition task or segmentation dataset with groundtruth.

The ALOI Dataset might be interesting for you: "ALOI is a color image collection of one-thousand small objects, recorded for scientific purposes. In order to capture the sensory variation in object recordings, we systematically varied viewing angle, illumination angle, and illumination color for each object, and additionally captured wide-baseline stereo images. We recorded over a hundred images of each object, yielding a total of 110,250 images for the collection."

The traffic sign recognition datset might also be interesting for you. IIRC there are also ground truth segmentation masks for the traffic signs.

In either cases you should be able to replace the background with whatever you want (if you want to make the task a abit more complicated).

Good luck with the recognition task (if it is still relevant).

---

I think CIFAR dataset might help you. It says that it involves labeled subsets of the 80 million tiny images dataset. Loading the dataset seems to be easy on Python, Matlab or C (binary version) via provided ways.

---

That is, I don't want to have to worry about pictures taken from different viewpoints or under different lighting conditions etc. Almost all the variability within a category should be due to the intrinsic variability of that category e.g. the category of cats would contain many different images because cats actually look different from one another, not because the images were produced under different conditions.

This is not very realistic. In fact, you will end up with a system that does not work in practice at all - or only in very constrained environments.

One such dataset I know (I did not care for such datasets) is the UIUC car datasets. This database contains images of side views of cars for use in evaluating object detection algorithms. (single-scale + multi-scale)

---

Try PASCAL, ImageNet & SUN. Caltech and MSRC are old and off.

You might want to read this paper as well: Unbiased Look at Dataset Bias - Torralba(MIT) & Efros (CMU): people.csail.mit.edu/torralba/publications/datasets_cvpr11.pdf‎