Package Exports

@tensorflow/tfjs-converter

This package does not declare an exports field, so the exports above have been automatically detected and optimized by JSPM instead. If any package subpath is missing, it is recommended to post an issue to the original package (@tensorflow/tfjs-converter) to support the "exports" field. If that is not possible, create a JSPM override to customize the exports field for this package.

Readme

Getting started

Tensorflow.js converter is an open source library to load a pretrained TensorFlow model into the browser and run inference through Tensorflow.js. It has two main pieces:

Coversion Python script, converts your Tensorflow SavedModel to web friendly format.
Javascript API, simple one line API for inference.

Dependencies

The python conversion script requires following packages:

  $ pip install tensorflow numpy absl-py protobuf

Usage

yarn add @tensorflow/tfjs-converter or npm install @tensorflow/tfjs-converter
Use the scripts/convert.py to convert your Tensorflow SavedModel.

$ python node_modules/@tensorflow/tfjs-converter/scripts/convert.py --saved_model_dir=/tmp/mobilenet/ --output_node_names='MobilenetV1/Predictions/Reshape_1' --output_graph=/tmp/mobilenet/web_model.pb --saved_model_tags=serve

Options	Description	Default value
saved_model_dir	Full path of the saved model directory
output_node_names	The names of the output nodes, comma separated
output_graph	Full path of the name for the output graph file
saved_model_tags	SavedModel Tags of the MetaGraphDef to load, in comma separated string format	serve

Outputs

This script would generate a collection of files, including model topology file, weight manifest file and weight files. In the above example, generated files are:

web_model.pb (model)
weights_manifest.json (weight manifest file)
group1-shard*of* (collection of weight files)

You need to have the model, weight manifest and weight files accessible through url. And the manifest and weight files should share the the same url path. For example:

  http://example.org/models/mobilenet/weights_manifest.json
  http://example.org/models/mobilenet/group1-shard1of2
  http://example.org/models/mobilenet/group1-shard2of2

Instantiate the TFModel class and run inference. Example

import {TFModel} from 'tfjs-converter';

const MODEL_FILE_URL = 'http://example.org/models/mobilenet/web_model.pb';
const WEIGHT_MANIFEST_FILE_URL = 'http://example.org/models/mobilenet/weights_manifest.json';

const model = new TFModel(MODEL_FILE_URL, WEIGHT_MANIFEST_FILE_URL);
const cat = document.getElementById('cat');
model.predict({input: dl.fromPixels(cat)}) // run the inference on your model.

Development

To build Tensorflow.js converter from source, we need to clone the project and prepare the dev environment:

$ git clone https://github.com/tensorflow/tfjs-converter.git
$ cd tfjs-converter
$ yarn prep # Installs dependencies.

We recommend using Visual Studio Code for development. Make sure to install TSLint VSCode extension and the npm clang-format 1.2.2 or later with the Clang-Format VSCode extension for auto-formatting.

Before submitting a pull request, make sure the code passes all the tests and is clean of lint errors:

$ yarn test
$ yarn lint

To run a subset of tests and/or on a specific browser:

$ yarn test --browsers=Chrome --grep='execute'
 
> ...
> Chrome 64.0.3282 (Linux 0.0.0): Executed 39 of 39 SUCCESS (0.129 secs / 0 secs)

To run the tests once and exit the karma process (helpful on Windows):

$ yarn test --single-run