pkg/utensor: add support to RIOT#12805
Merged
fjmolinas merged 3 commits intoRIOT-OS:masterfrom Jan 7, 2020
Merged
Conversation
aabadie
added
Type: new feature
aabadie
force-pushed
the
pr/pkg/utensor
branch
3 times, most recently
from
e182301 to
822c18d
Compare
Contributor
|
force-pushing a lot... |
Contributor
Author
Indeed, I'm trying to fix the build issues reported by Murdock. Since no one reviewed yet, I think that's ok :) |
aabadie
added
CI: ready for build
aabadie
changed the title
josephnoir
reviewed
Jan 6, 2020
Contributor
josephnoir
left a comment
josephnoir
left a comment
There was a problem hiding this comment.
Hi, sorry for taking so long. I left a few comments regarding the C++ code and checked that it works on a board here. As far as I can tell the Makefile stuff looks ok.
aabadie
force-pushed
the
pr/pkg/utensor
branch
from
822c18d to
ae8e6ff
Compare
Contributor
Author
|
Thanks for your initial review @josephnoir. I addressed all your comments. I also noticed that the utensor-cli (used to generated the model/weights ccp files) has been updated but the version on pypi is now broken so I opened uTensor/utensor_cgen#97 to provide a fix. |
josephnoir
approved these changes
Jan 7, 2020
aabadie
removed
the
CI: ready for build
aabadie
force-pushed
the
pr/pkg/utensor
branch
from
ae8e6ff to
8a9ded9
Compare
aabadie
added
CI: run tests
Contributor
Author
|
Since the test application provides an automatic test script, I ran Murdock with |
aabadie
added
CI: ready for build
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Contribution description
This PR adapts the uTensor library to RIOT as a package.
The sample application provided in the tests is just an adaption of the example application given in this blog post.
The application is demonstrating how to load a model (model structure + weights) generated from a trained model to determine a digit in an image.
The model is trained using TensorFlow on the MNIST dataset.
The digit image is loaded in the firmware using the recently introduced blob mechanism. There is a helper Python script added to be able to update the digit binary file easily: this script load the test data from the MNIST dataset and, given the index in the list of image, save the image on disk and display its content. This is handy for the user to be able to compare with the value detected by the firmware.
The model files (c++) are automatically generated by the utensor-cli tool and stored in the
modelssubdirectory of the application. These C++ files (model description + weights) are loaded in the application as an external module.The resulting firmware is quite huge, ~180kiB ROM size, and the package requires C++ support, so a few boards (AVR, MSP430, hifive) cannot be used. It should work on MIPS :)
Testing procedure
The following command should succeed:
Issues/PRs references
None