From: Rosie.Sachdev@sceiwa.tech
Sent: Tuesday, 3rd March 2009 3:51 PM
To: Monti.Amal@sceiwa.tech
Subject: RE: OMI Image Generation Bug

Hi Monti,

All jokes aside, I’ve just made a bug for it: “The Featherless Biped Problem”. I’ve added the training plan, if you could add the import/export sets. Hopefully we can track down where the modelling went a little wonky.

Thanks,
Rosie Sachdev

From: Monti.Amal@sceiwa.tech
Sent: Tuesday, 3rd March 2009 3:26 PM
To: Rosie.Sachdev@sceiwa.tech
Subject: OMI Image Generation Bug

Hello Rosie,

I have completed the most recent round of training oversight. Unfortunately the outcomes were not what we hoped for. While the OMI had a high success rate identifying our library of animal images, the images generated from the tags in round 2 came out very garbled.

For example, OMI was given a picture of a dog and tagged it {4 legs, tail, 2 ears, 2 eyes}. Given its own tag set it produced the below images, seemingly merging four duck legs, two bunny ears and a lizard tail into a ‘dog’.

<OMI_OutputImage1.png>
<OMI_OutputImage2.png>

Kind Regards,
Monti Amal