Machine Learning Classification of different jewelry items

Machine Learning Classification of different jewelry items

Cancelled

Job Description

We have different jewelry images. These images need to be recognized (= which item) the angle needs to be recognized and then patterns need to be found *across different items* so they can be classified into the same classes.

Classes are probably most likely to work using design elements and principles of design: color (centers), symmetry, shapes, curves, lots of detail, minimalism etc etc.

Where the real problematic part comes is in classifying those correctly. It would probably be neural networks and probably be supervised machine learning -- though no idea how big the training set would have to be.

For example: would

https://scontent-a-ams.xx.fbcdn.net/hphotos-ash3/563615_10152094028604375_1690065262_n.jpg

be classified in the same class as

https://scontent-b-ams.xx.fbcdn.net/hphotos-frc3/1479203_10152095677074375_1058513797_n.jpg ?? And why?

It's obvious "art deco" has connections to modern minimalistic design but how much connection?

And this one
https://scontent-b-ams.xx.fbcdn.net/hphotos-prn2/1526263_10152094028264375_585150331_n.jpg

vs

https://scontent-a-ams.xx.fbcdn.net/hphotos-prn1/64198_10152085497924375_51424636_n.jpg

vs

https://scontent-a-ams.xx.fbcdn.net/hphotos-prn2/1505067_10152084972324375_2090677708_n.jpg

They're all colorful, high level of details. A ML algorithm could potentially classify them in the same group.. but the last one is more contemporary artistic.. the butterfly is traditional and naturalistic and the bracelet is something in between I guess.

A human may pick up on PART of a design and feel that that is the most important, wanting more of it (so her classification would focus on that part) or giving it more emphasis than an algorithm would perhaps. As well as certain emotional values.

Not so easy to execute

Other open jobs by this client