Question: VisionBonnet & TensorFlow

google / aiyprojects-raspbian

API libraries, samples, and system images for AIY Projects (Voice Kit and Vision Kit)

https://aiyprojects.withgoogle.com/

Apache License 2.0

1.63k stars 694 forks source link

Question: VisionBonnet & TensorFlow #566

Closed edrdo closed 5 years ago

edrdo commented 5 years ago

I have a few questions regarding the AIY Vision Kit in particular regarding VisionBonnet and also TensorFlow - sorry if some most of these go beyond strict development issues, but here seems to be the better place to ask:

Does the VisionBonnet chip use TensorFlow?
Either way, is it possible to convert the models available at the AIY projects for use with TensorFlow in other settings (e.g. using the TF API directly) ?
Is the standalone VisionBonnet chip available for sale? Also, how does it releate to the upcoming Edge TPU Devices ?

dmitriykovalev commented 5 years ago

VisionBonnet is based on Myrad2450 chip.

There is a special binary format to represent compute graphs. VisionBonnet Python API is designed to run these files. Some compute graphs can be converted from TensorFlow by using our compiler.

EdgeTPU devices are based on a different chip (Edge TPU) designed by Google.

edrdo commented 5 years ago

Thanks for the answers.

Some compute graphs can be converted from TensorFlow by using our compiler.

I see, but I was looking for a way to accomplish the reverse operation, i.e., to convert a Vision Kit model to TensorFlow format. In particular I was interested in doing so for the Nature Explorer model.

weiran-work commented 5 years ago

@edrdo

Unfortunately, there's no method to do the reverse, during the conversion, some information is lost.