- Notifications
You must be signed in to change notification settings - Fork0
License
hpssjellis/tensorflowjs-to-arduino-for-tinymljs
Folders and files
| Name | Name | Last commit message | Last commit date | |
|---|---|---|---|---|
Repository files navigation
Not presently working on windows, hopefully that changes.Actually presently I can't even get this working on Ubuntu. Try the ipythonhttps://colab.research.google.com/drive/1OgCcKhklL3EH_SdWHdtlb5dbtYvjGQnn?usp=sharing or gitpod optionshttps://gitpod.io/#github.com/hpssjellis/Gitpod-auto-tensorflowJS-to-arduino
- Go to the location you downloaded the model .json and model.bin files fromhttps://hpssjellis.github.io/tinyMLjs/public/index.html
- Instal venv
sudo apt install python3.12-venvso you can work in an environment and not have other things mess it up - Install a virtual environment
python3 -m venv myenv10 - Activate that environment (note it has a folder you can stay out of )
source myenv10/bin/activateor on windowsmyenv10\scripts\activate pip install tensorflowpip install tensorflowjspip install tensorRTtflite_convert --helptensorflowjs_converter --help10.tensorflowjs_converter --input_format=tfjs_layers_model --output_format=keras_saved_model ./model.json ./Convert tfjs file to kerastflite_convert --keras_model_file ./ --output_file ./model.tfliteConvert Keras file to tflite filexxd -i model.tflite model.hConvert tflite file to a c-header file (This needs xxd installed, several ways to do this also can do it from a web page)
Non of the above work for me that is why I have the other options
- use python notebookshttps://colab.research.google.com/drive/1OgCcKhklL3EH_SdWHdtlb5dbtYvjGQnn?usp=sharing (run both sketches then upload your files and run the last sketch again)
- This repohttps://github.com/hpssjellis/Gitpod-auto-tensorflowJS-to-arduino and run the gitpod which has a file that does the conversions. Basically it installs the above and then you can run the commands or a bash file I have ready to do the conversions for you. The autoloading gitpod is herehttps://gitpod.io/#github.com/hpssjellis/Gitpod-auto-tensorflowJS-to-arduino.
.
.
This github repository is athttps://github.com/hpssjellis/tensorflowjs-to-arduino-for-tinymljs
if you want to load it as a Gitpod clickhttps://gitpod.io/#github.com/hpssjellis/tensorflowjs-to-arduino-for-tinymljs
Click the above link to load the gitpod (docker in the browser) which installs all needed files
You can use the test-with folder to drag amodel.json file with it's shard .bin filemodel.weights.bin to the main folder
Look at the code in thea01-convert-tfjs-arduino.sh and then run it
Then run./a01-convert-tfjs-arduino.sh
Gotchas When making your own files themodel.json file is made with a link to themodel.weights.bin file, if you change the name of the binary file the model.json fle will not link to it properly
I assume python is installed probably best to have Python3 installed.
pip install tensorflowjspython -m site --user-baseto the above reply add
\bin\tensorflowjs_converter -hThen run the commands for your files which are in the a01-convert-tfjs-arduino.sh bash file
tensorflowjs_converter --input_format=tfjs_layers_model --output_format=keras_saved_model ./model.json ./tflite_convert --keras_model_file ./ --output_file ./model.tflitexxd -i model.tflite model.hThen you can load your model.tflite file onto thehttps://netron.app/ website to visualize it and then add the model.h file into your arduino machine learning code as it;s own include file.
See the Arduino ready library athttps://github.com/hpssjellis/RocksettaTinyML download the zip file and install it into the arduino ide using the normal zip file libary upload method.sketch --> include library --> add .zip file
Note: If the above commands don't work you can always try the python code below.
- convert from tensorflowJS to Keras
import tensorflowjs as tfjs# Define the pathsinput_format = "tfjs_layers_model"output_format = "keras_saved_model"input_model_json = "./model.json"output_dir = "./"# Convert the modeltfjs.converters.save_keras_model(input_model_json, output_dir, input_format, output_format)Then use tensorflow lite converter to convert the Keras file into tensorflow Lite (TFLITE)
import tensorflow as tf# Define the pathskeras_model_file = "./model" # Make sure the model file has the .h5 extensionoutput_file = "./model.tflite"# Convert the model to TensorFlow Liteconverter = tf.lite.TFLiteConverter.from_keras_model_file(keras_model_file)tflite_model = converter.convert()# Save the TensorFlow Lite model to a filewith open(output_file, 'wb') as f: f.write(tflite_model)If you have install ability then install the xxd application
sudo apt-get install xxdand run the command
xxd -i model.tflite model.hIf you don't have admin access you can try using the online xxd -1 utility herehttps://hpssjellis.github.io/tinyMLjs/public/convert/xxd-i.html
Upload your tFLITE file and get the web to convert it into a c-header model.h file ready to run on a micro-controler with an appropriate sketch.
python.exe -m pip install --upgrade pip
pip3 install --upgrade pip
python -m venv myenv2
myenv2\scripts\activate
pip3 install tensorflowjs
pip3 install tensorflow==2.15.0
pip3 install tensorflow-hub
pip3 install netron "dask[delayed]"
$env:TF_ENABLE_ONEDNN_OPTS=0
tflite_convert --helptensorflowjs_converter --help
xxd --help
https://sourceforge.net/projects/xxd-for-windows/
in power shell try
Format-Hex '.\your-file-name'
this set works
pip list--------------------------------- ---------absl-py 2.1.0argon2-cffi 21.1.0astroid 2.7.3astunparse 1.6.3attrs 21.2.0autopep8 1.5.7backcall 0.2.0backports.entry-points-selectable 1.1.0bandit 1.7.0bleach 4.1.0cached-property 1.5.2cachetools 5.3.3certifi 2021.5.30cffi 1.14.6charset-normalizer 2.0.4chex 0.1.7click 8.1.7cloudpickle 3.0.0colorama 0.4.4cryptography 3.4.8dask 2023.5.0debugpy 1.4.3decorator 5.1.0defusedxml 0.7.1distlib 0.3.2dm-tree 0.1.8docutils 0.17.1entrypoints 0.3etils 1.3.0filelock 3.0.12flake8 3.9.2flatbuffers 24.3.25flax 0.7.2fsspec 2024.5.0gast 0.4.0gitdb 4.0.7GitPython 3.1.18google-auth 2.29.0google-auth-oauthlib 1.0.0google-pasta 0.2.0grpcio 1.63.0h5py 3.11.0idna 3.2importlib_metadata 7.1.0importlib_resources 6.4.0ipykernel 6.4.1ipython 7.27.0ipython-genutils 0.2.0isort 5.9.3jax 0.4.13jaxlib 0.4.13jedi 0.18.0jeepney 0.7.1Jinja2 3.0.1jsonschema 3.2.0jupyter-client 7.0.2jupyter-core 4.7.1jupyterlab-pygments 0.1.2keras 2.13.1keyring 23.2.1lazy-object-proxy 1.6.0libclang 18.1.1locket 1.0.0Markdown 3.6markdown-it-py 3.0.0MarkupSafe 2.1.5matplotlib-inline 0.1.3mccabe 0.6.1mdurl 0.1.2mistune 0.8.4ml-dtypes 0.2.0msgpack 1.0.8mypy 0.910mypy-extensions 0.4.3nbclient 0.5.4nbconvert 6.1.0nbformat 5.1.3nest-asyncio 1.5.1netron 7.6.6notebook 6.4.3numpy 1.24.3oauthlib 3.2.2opt-einsum 3.3.0optax 0.1.8orbax-checkpoint 0.2.3packaging 23.2pandas 2.0.3pandocfilters 1.4.3parso 0.8.2partd 1.4.1pbr 5.6.0pep8 1.7.1pexpect 4.8.0pickleshare 0.7.5pip 24.0pipenv 2021.5.29pkginfo 1.7.1platformdirs 2.3.0prometheus-client 0.11.0prompt-toolkit 3.0.20protobuf 4.25.3ptyprocess 0.7.0pyasn1 0.6.0pyasn1_modules 0.4.0pycodestyle 2.7.0pycparser 2.20pydocstyle 6.1.1pyflakes 2.3.1Pygments 2.18.0pylama 7.7.1pylint 2.10.2pyparsing 2.4.7pyrsistent 0.18.0python-dateutil 2.8.2pytz 2024.1PyYAML 5.4.1pyzmq 22.2.1readme-renderer 29.0requests 2.26.0requests-oauthlib 2.0.0requests-toolbelt 0.9.1rfc3986 1.5.0rich 13.7.1rope 0.19.0rsa 4.9scipy 1.10.1SecretStorage 3.3.1Send2Trash 1.8.0setuptools 58.0.4six 1.16.0smmap 4.0.0snowballstemmer 2.1.0stevedore 3.4.0tensorboard 2.13.0tensorboard-data-server 0.7.2tensorflow 2.13.1tensorflow-decision-forests 1.5.0tensorflow-estimator 2.13.0tensorflow-hub 0.16.1tensorflow-io-gcs-filesystem 0.34.0tensorflowjs 4.19.0tensorstore 0.1.45termcolor 2.4.0terminado 0.12.1testpath 0.5.0tf-keras 2.15.0toml 0.10.2toolz 0.12.1tornado 6.1tqdm 4.62.2traitlets 5.1.0twine 3.4.2typing_extensions 4.5.0tzdata 2024.1urllib3 1.26.6virtualenv 20.7.2virtualenv-clone 0.5.7wcwidth 0.2.5webencodings 0.5.1Werkzeug 3.0.3wheel 0.37.0wrapt 1.12.1wurlitzer 3.1.0zipp 3.5.0More attempts I think it is my python version is either to recent or not old enough
new instructions
python.exe -m pip install --upgrade pip
python -m venv myenv10myenv10\scripts\activate
try the newest versions but if that doesn't work these versions will work.pip install tensorflow==2.13.1pip install tensorflow-decision-forests==1.4.0pip install tensorflowjs==4.19.0 --no-deps
pip install tensorflow-decision-forests==1.8.0 --no-depspip install tensorflow-decision-forests==1.5.0
--ignore-installed tensorflow_decision_forests tensorflow tensorflow-io-gcs-filesystem tensorstore
tflite_convert --helptensorflowjs_converter --help
pip install tensorflowjs --no-deps
pip install tensorflowjs==4.19.0
pip install tensorflow==2.13.1pip install tensorflowjs==4.19.0
tflite_convert --helptensorflowjs_converter --help
#!/bin/bash
tensorflowjs_converter --input_format=tfjs_layers_model --output_format=keras_saved_model ./model.json ./tflite_convert --keras_model_file ./ --output_file ./model.tflitexxd -i model.tflite model.h
python.exe -m pip install --upgrade pippip3 install --upgrade pip
python -m venv myenv2myenv2\scripts\activate
pip3 install tensorflowjspip3 install tensorflow==2.15.0pip3 install tensorflow-hubpip3 install netron "dask[delayed]"
$env:TF_ENABLE_ONEDNN_OPTS=0
tflite_convert --helptensorflowjs_converter --help
xxd --help
https://sourceforge.net/projects/xxd-for-windows/
in power shell try
Format-Hex '.\your-file-name'
About
Resources
License
Uh oh!
There was an error while loading.Please reload this page.
Stars
Watchers
Forks
Releases
Packages0
Uh oh!
There was an error while loading.Please reload this page.