The link below, quoted by Intel support staff member Leonardo, explains how objects are detected using bounding boxes and confidence percentages.
Leonardo added, "You can only use the images that are described in the website [above]."
Thanks Marty, but that's not what I'd like to know.
Based on the link you posted here: Developing Apps with Object Library | Intel® Software , I need to setup a training folder to use YOR.
There are more details on what's supposed to be in that folder in the headers (see or_video_module_impl.h:86 from the SDK).
I'd like Intel to share the training folder they used to train the Object Library w/ the classes Leonardo mentioned.
I would imagine that the training folder will not be shareable, but there is a page that describes in text how training images are made.
Basically, you take pictures of an object from different angles and in different lighting conditions and backgrounds and then provide that image set to the training application so that it can learn how to recognize that class of object.
I'm hoping they will be nice to the developers who use their platform and agree to share it.
It's not like it's some super secret algorithm, more like an example for their SDK.
This message was posted on behalf of Intel Corporation
We were able to investigate this request on our end with our engineering team and we got the following:
"The training set of the object recognition classifier (Not the YOR Tool) are part of the closed IP and cannot be shared. We have no other reference training sets for the YOR tool to provide. If you would like to maintain the 42 objects of the Object Recognition feature while adding new objects with the YOR Tool, that’s not possible.
The goal of the YOR Tool is to create a totally new classifier from the training images set that the user has."
We hope that this answers your inquiry and we apologize for any inconvenience this may cause.
Thanks for investigating and getting back to me.
That's very unfortunate and doesn't make much sense to me.