Flite software package and tensorflow

[jvarkey]

Hi Lukas,
Could you let me know the details of how Flite software is linked with tensorflow? I have followed your directions and at this time I can see what my camera is seeing through localhost/cam.jpg. However, I did not see directions in the article how to install Flite so that it can speak the objects that tensorflow is outputting. If you can provide directions as to the installation and how the two programs can be linked, it will be great.
Also, thank you so much for sharing this work and the article.

Your sincerely,
John Varkey

[fmacrae]

Hi jvarkey,

I've got to the same stage too. All the installation went fine as per Lukas' guide but I can't get any of the Python programs or run_and_parse_inception.sh to work. It's like I'm missing a pre-requisite as the script seems to want a ~/tensorflow directory. The tensorflow-0.11.0-cp27-none-linux_armv7l.whl installs tensorflow into /usr/local/lib/python2.7/dist-packages/tensorflow. Tried creating symbolic link to it using
cd ~
ln -s /usr/local/lib/python2.7/dist-packages/tensorflow tensorflow

But no luck. I'll add to this issue if I figure it out. Maybe we can contribute a quickstart guide to the project if we figure it out before Lukas gets back to us ;)

Regards,

Finlay

[jvarkey]

Hi Finlay,

We got it working but had to follow both the links below which were mentioned by Lukas in the Oreilly article. The first one to install tensorflow and the second one for label_image
Compilation took about three hours. Then we modified run_and_parse_inception.py to point to where tensorflow was installed, as seen below. At this point, if you run image.py the robot will speak what it is seeing.

https://github.com/tensorflow/tensorflow/tree/master/tensorflow/contrib/makefile#raspberry-pi
https://github.com/tensorflow/tensorflow/tree/master/tensorflow/contrib/pi_examples

( cd /home/pi/tensorflow-master && tensorflow/contrib/pi_examples/label_image/gen/bin/label_image --image=../robot/$1) &> ../robot/raw.output
tail -5 raw.output | cut -d"]" -f2 | cut -d"(" -f1 > $2

When I previewed the comment, the style looks odd.

Regards,
JV

[fmacrae]

You guys are AWESOME!!! jvarkey, I just got my rasperry pi tensorflow working due to your help. Think we need to get a how to guide wiki going as even some of the electrical stuff is somewhat tricky. I nearly plugged the sonars into the GPIO running at 5V. Luckily I realised in time that I'd need a voltage divider to lower the echo down to 3.3V. Sure there's been a few fried Rasperry Pis caused by that.