Re:Indigenous datasets: A listing to help AI perform better ...
by johnny313 on Tuesday March 12, 2019 @04:13PM
(#19372500)
This reminds me of a paper put out by a few researchers at Google Brain: `No Classification without Representation: Assessing Geodiversity Issues in Open Data Sets for the Developing World` [0]
The take away was that existing open image datasets are biased toward western contexts (eg what a wedding looks like), leading to low performance when applied in non-western contexts.
Re:Indigenous datasets: A listing to help AI perform better ...
by sairahul82 on Tuesday March 12, 2019 @05:10PM
(#19373067)
https://tdil-dc.in/index.php?lang=en has good some good datasets. Only problem is you need to agree the t&s and fax the document to Delhi to get the access. They supposedly will send the DVD. I don't understand why they are doing. If Indian government wants to foster the research, they should put them in public domain.
Here are few more datasets which I am aware of and can be added to the list:
- Dataset for Indian National Rupee https://cvit.iiit.ac.in/research/projects/cvit-projects/curr...
- City-scale Road Audit using Deep Learning dataset https://cvit.iiit.ac.in/research/projects/cvit-projects/city...
- Multi domain corpus for sentimental analysis (Telugu dataset, download link isn't working probably email for download)
The take away was that existing open image datasets are biased toward western contexts (eg what a wedding looks like), leading to low performance when applied in non-western contexts.
[0] https://arxiv.org/abs/1711.08536