Collecting large image databases

How to

Take a look at Greg Griffin report on collecting Caltech 256

Scripts

Rob Fergus has scripts to do this

Is it legal?

Copyright is an issue that often comes up.

The provision of `Fair use' (see Wikipedia page on fair use) exempts most scientific uses from the need of requesting a permission from the holder of copyright. This is because we:

• do not make money using the pictures
• do not decrease the value of the pictures for the person who owns them
• add to the content (by performing computations on them and deriving descriptors, also by publishing theory and experiments on them)
• we only use a fraction of the work of a given copyright holder
• our work benefits the public and is given out for free
• we use the images because they represent a sample of the real world, the `artistic' and `creative' aspects involved in image taking are immaterial (and in fact detrimental) to our work
• we typically do not show the images at all, just numbers extracted from them

When we post our datasets we should be careful to indicate that we do not own right to the pictures and anyone using our collections is responsible of verifying that they are doing so legally.

Discussion on fair use:

• Wikipedia page on fair use
• http://www.copyright.gov/fls/fl102.html
• http://fairuse.stanford.edu/Copyright_and_Fair_Use_Overview/chapter9/9-b.html
• http://www.copyright.iupui.edu/highered.htm