Questions

[Иван Танкоев]

Hello, I'm researcher from Trento. I'm interesting in your project.I
have a couple of questions.
1. Are you going to provide some API like in Flickr.
2. Do you create some kind of visual signature for each concept in you
ImageWordNet ?

Thank you in advance,
Ivan

[rula....@coloplastmena.com]

Hello

Actually I wasn't able to download imagenet and use it. I didn't get the
approval to download it. Therefore I am sorry I cannot answer your questions
unless they allow me to get the data.

Cheers
Rula

[Alexander Sorokin]

Hi Ivan

1. There's an API already:
http://www.image-net.org/download-API

2. Visual signatures would be awesome to ensure that the Image URL
still maps to the annotated image. It would be great to have a unique
image ID : URL?visual_signature=ZXWSDASD as some sort of a standard.
Then someone else can provide other visual features to these images.

Disclaimer: I'm not running the project.

Cheers!
Alex

[Jia Deng]

Hi Ivan,

Thanks for your interest in ImageNet. As Alex as pointed out, we do
have a set of APIs ( and thank you, Alex ). By "visual signature", do
you mean a set of representative images for each concept or some
representation in some feature space ( e.g. an average image )?

Thanks,
Jia Deng

[Jia Deng]

Hi Alex,

Currently for each ImageNet image, there is indeed a global id, in the
form of XXXXX_YYYY where XXXXX is the synset offset and YYYY is an
integer. If that's what you were referring to, we will be able to
provide an API to map an image id to an URL, or to some other
information. Or did you mean something else? Thanks.

Jia Deng

[Alexander Sorokin]

Hi Jia,

I was also saying that it's beneficial to provide a visual signature
of the image. I.e. map XXXXX_YYYY -> hash(GIST(img)) . In this way if
the owner of the URL decides to replace the image with a new one, the
users of the data will know that the image at that URL is no longer
valid for that annotation.

Alex

[Hao]

Hi Alex,

That's a good suggestion. We would like to consider it as a feature.

Thanks,
Hao

--
Thanks,
Hao

[Fei-Fei Li]

Alex,
Indeed, I agree with you. Providing such features would be very useful.
And this is what we plan to do in the Spring Release (around April).
We'll provide feature vectors to the images in ImageNet. We plan to
start with Bag of Words histogram of a fixed dictionary. But more should
come. :)

Great suggestion!
Fei-Fei

---------------------------------------------
Li, Fei-Fei Ph.D.
(publish under L. Fei-Fei)
Assistant Professor
Computer Science Dept.
Stanford University
353 Serra Mall, 2A Room 246
Stanford, CA 94305-9025

Tel: (650)725-3860
Website: http://vision.stanford.edu
---------------------------------------------