search

robmarkcole / coral-pi-rest-server

Perform inferencing of tensorflow-lite models on an RPi with acceleration from Coral USB stick
https://coral.ai/products/accelerator
MIT License
67 stars 20 forks source link

Added min_confidence. Fixed bounding box issue. #75

Closed sstratoti closed 1 year ago

sstratoti commented 1 year ago

This should fix #58 - It'll allow you to set a min_confidence in the form post similar to how deepstack operates. I also noticed that my bounding boxes were all out of whack, similar to #62 . I think @robmarkcole was right on the money with the resize to 300x300.

Following this example: https://github.com/google-coral/pycoral/blob/master/examples/detect_image.py

I reworked it so that it didn't resize the image prior to calling interpreter.invoke(). I also borrowed some code to write out how much processing time each run takes, but I commented it out so as not to spam the log.

I also noticed that you could reference bbox.ymin in the pycoral docs, so I swapped out the bbox[0] so that its a little more clear.

Lastly, I switched out the print for the global interpreter to use a logging.debug instead. This'll print it out to the docker container.

BTW - once a new release is made in this repo, I can update https://github.com/sstratoti/docker-coral-rest-server Dockerfile which I forked from another user and have since updated with some fixes. This is working pretty great so far!

robmarkcole commented 1 year ago

Thanks @sstratoti !

sstratoti commented 1 year ago

Thanks, https://github.com/sstratoti/docker-coral-rest-server is now updated to use 2.2. Don't know if I'll be able to maintain it quickly moving forward, but if anyone wants to submit a PR I can merge it real quick.