Jump to content

WOW! I gotta build one of these!


G+_Mark Olson
 Share

Recommended Posts

Oh ok. It uses some cloud AI to process the picture and then send back ANOTHER picture (which might not look anything like the original) to print.

 

It wasn't very clear to me from the article. I knew nothing about the AI and saw an arduino, and wondered how the arduino could do all of this.

 

Something similar, like OCR (optical character recognition), on a raspberry pi 3B+ takes like 10 seconds. Cannot imagine it on an arduino. Cloud processing to the rescue.

Link to comment
Share on other sites

I don't know everything about it, but I think there is a Raspberry Pi for the processing. I don't see an Arduino in the github repository "schematic"; I think that board in the picture is a Raspberry Pi. After I get a chance to look at the code I hope to figure out if it is doing the processing in the cloud.

Link to comment
Share on other sites

There is some Arduino code that seems to be designed to turn on/off or blink LEDs during certain time periods, but I don't think it is actually used. The LED is wired into the Raspberry Pi. Also, there is a website that allows you to test the AI code online against a picture but so far it looks like the AI code (based on Google TensorFlow) would run directly on the Raspberry Pi.

Also, the hardware part of this project seems to be based on this AdaFruit project: learn.adafruit.com - Overview | Instant Camera using Raspberry Pi and Thermal Printer | Adafruit Learning System

Link to comment
Share on other sites

Oops. You're right. It does look like a raspberry pi in the picture. I just looked at the github, saw arduino and moved on. Had I bothered to actually hit the scroll button ONCE, I would have seen some raspberry pi stuff. How embarrassing! :) From what I can tell, it looks like a google open source library tensor flow for the AI. Still, I wonder about what the response time is to process the image. Care to try it out? ;)

Link to comment
Share on other sites

Jeff Gros Oh yeah, I am definitely trying it out! I got a Raspberry Pi 3B+, RPI camera V2, and Adafruit Tiny TTL Thermal Printer. Also I looked into some other things I can do, like use AI to put a caption on the picture (https://github.com/boluoyu/ImageCaption) or generate silhouettes (https://github.com/bonlime/keras-deeplab-v3-plus). Also there is usually a layer in the neural net that basically derives edges; that is a fun picture in many cases. The doodle one reminds me of the Terry Pratchet Diskworld cameras that use an imp to draw a picture.

Link to comment
Share on other sites

I'm interested in the printer.

 

Not sure what the printer actually is, but looking at the python code, I can see it uses the ESC/POS protocol. That doesn't necessarily mean it is an Epson under the hood though, as other brands support the protocol as well.

 

Can you peek under the hood and see if there's any identifying marks inside?

Link to comment
Share on other sites

Jeff Gros I guess I actually got the Nano Thermal Receipt Printer – TTL Serial. You can see details (https://blog.adafruit.com/2015/12/21/new-products-tiny-thermal-receipt-printer-nano-thermal-receipt-printer-thermal-receipt-printer-guts-thermal-paper-roll-33-long-thermal-paper-roll-16-long/) about halfway down the page. I don't see any identifying marks in those pictures (much better than anything I would do) or looking at my printer. My guess: not Epson.

Link to comment
Share on other sites

 Share

×
×
  • Create New...