3
$\begingroup$

I came across a paper that is trying to classify and detect object(s) present in what they call a natural scene data-set. They have images containing objects like cars, bikes and people.

Can anyone tell me how this is different from other data-sets that contain the images of cars/bikes/people?

The site with the data-sets I am interested in is: Natural Scene Database.
For comparison, the Caltech Computational Vision data-set also contains cars and bikes from the rear and side view (with less background).

I want to know what is so special about the natural image data-set compared to the standard data-sets containing various viewpoints of cars/bikes from these set of images? Why would this natural image data-set be a bigger callenge in object detection?

Improve this question
$\endgroup$
3
  • $\begingroup$ This question appears either as being extremely broad or unclear. It could be helpful to specify what kind of an answer you expect. You should be able to edit your question. $\endgroup$
    – user7358
    Feb 26, 2014 at 19:10
  • 1
    $\begingroup$ Hey, the site you linked to seems to be a cars/bikes dataset. I don't understand if you're claiming the datasets on your link are Natural Scene Datasets, and you want to compare them to some other bikes/people/cars datasets, or if those are the bikes/cars datasets you want us to compare the "Natural Scenes Datasets" to $\endgroup$
    – penelope
    Feb 28, 2014 at 9:45
  • $\begingroup$ @penelope The data-set used in that website is that of natural scenes.It is described in the papers that are available in that website.On the other hand if you see this website,Caltech Computational Vision,you will find datasets of cars and bikes in the rear and side view only with less background.My question is,How is it more challenging to detect objects in the Graz University dataset compared to that of the website that I have referred to in this section.Thanks and regards $\endgroup$
    – logamadi
    Feb 28, 2014 at 10:12

3 Answers 3

Reset to default
2
$\begingroup$

From one perspective, not much. However, biological perceptual systems have evolved in response to physical properties of natural environments. And such environments possess certain characteristics, such as Laplacian distribution shaped histograms. Therefore natural images are grouped and analyzed together.

Improve this answer
$\endgroup$
0
2
$\begingroup$

A natural image is like a photo, as opposed to a synthetic or artificial image which is like a cartoon or line art, that typically has sharp edges and flat areas of color.

It's of interest to people who are compressing digital images: natural images compress better with JPEG while synthetic ones compress better using lossless formats such as PNG or GIF.

The distinction between these two types is blurring these days, as technology for photorealistic image generation has developed.

Improve this answer
$\endgroup$
1
$\begingroup$

It's mostly an historical heritage. What the authors mean is that none of the pictures were:

  • computer generated
  • shot in controlled conditions (e.g., photobooth, lab...)
  • faked or set up in any way.

Today, the trending term for this is in the wild. This does not mean that the images will bite you, but that the images come from standard customer/everyday cameras instead.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.