Awesome

Pre-trained TensorFlow.js models

This repository hosts a set of pre-trained models that have been ported to TensorFlow.js.

The models are hosted on NPM and unpkg so they can be used in any project out of the box. They can be used directly or used in a transfer learning setting with TensorFlow.js.

To find out about APIs for models, look at the README in each of the respective directories. In general, we try to hide tensors so the API can be used by non-machine learning experts.

For those interested in contributing a model, please file a GitHub issue on tfjs to gauge interest. We are trying to add models that complement the existing set of models and can be used as building blocks in other apps.

Models

<table style="max-width:100%;table-layout:auto;"> <tr style="text-align:center;"> <th>Type</th> <th>Model</th> <th>Demo</th> <th>Details</th> <th>Install</th> </tr>   <tr> <td rowspan="12"><b>Images</b></td> <td rowspan="2"><b><a style="white-space:nowrap; display:inline-block;" href="./mobilenet"><div style='vertical-align:middle; display:inline;'>MobileNet</div></a></b></td> <td><a href="https://storage.googleapis.com/tfjs-models/demos/mobilenet/index.html">live</a></td> <td rowspan="2">Classify images with labels from the <a href="http://www.image-net.org/">ImageNet database</a>.</td> <td rowspan="2"><code>npm i @tensorflow-models/mobilenet</code></td> </tr> <tr> <td><a href="./mobilenet/demo/index.html">source</a></td> </tr>  <tr> <td rowspan="2"><b><a style="white-space:nowrap; display:inline-block;" href="./hand-pose-detection"><div style='vertical-align:middle; display:inline;'>Hand</div></a></b></td> <td><a href="https://storage.googleapis.com/tfjs-models/demos/hand-pose-detection/index.html?model=mediapipe_hands">live</a></td> <td rowspan="2">Real-time hand pose detection in the browser using TensorFlow.js.</td> <td rowspan="2"><code>npm i @tensorflow-models/hand-pose-detection</code></td> </tr> <tr> <td><a href="./hand-pose-detection/demos/live_video/index.html">source</a></td> </tr>  <tr> <td rowspan="2"><b><a style="white-space:nowrap; display:inline-block;" href="./pose-detection"><div style='vertical-align:middle; display:inline;'>Pose</div></a></b></td> <td><a href="https://storage.googleapis.com/tfjs-models/demos/pose-detection/index.html?model=movenet">live</a></td> <td rowspan="2">An API for real-time human pose detection in the browser.</td> <td rowspan="2"><code>npm i @tensorflow-models/pose-detection</code></td> </tr> <tr> <td><a href="./pose-detection/demos/live_video/index.html">source</a></td> </tr>  <tr> <td rowspan="2"><b><a style="white-space:nowrap; display:inline-block;" href="./coco-ssd"><div style='vertical-align:middle; display:inline;'>Coco SSD</div></a></b></td> <td><a href=""></a></td> <td rowspan="2">Object detection model that aims to localize and identify multiple objects in a single image. Based on the <a href="https://github.com/tensorflow/models/blob/master/research/object_detection/README.md">TensorFlow object detection API</a>.</td> <td rowspan="2"><code>npm i @tensorflow-models/coco-ssd</code></td> </tr> <tr> <td><a href="./coco-ssd/demo">source</a></td> </tr> <td rowspan="2"><b><a style="white-space:nowrap; display:inline-block;" href="./deeplab"><div style='vertical-align:middle; display:inline;'>DeepLab v3</div></a></b></td> <td><a href=""></a></td> <td rowspan="2">Semantic segmentation</td> <td rowspan="2"><code>npm i @tensorflow-models/deeplab</code></td> </tr> <tr> <td><a href="./deeplab/demo">source</a></td> </tr>  <tr> <td rowspan="2"><b><a style="white-space:nowrap; display:inline-block;" href="./face-landmarks-detection"><div style='vertical-align:middle; display:inline;'>Face Landmark Detection</div></a></b></td> <td><a href="https://storage.googleapis.com/tfjs-models/demos/face-landmarks-detection/index.html?model=mediapipe_face_mesh">live</a></td> <td rowspan="2">Real-time 3D facial landmarks detection to infer the approximate surface geometry of a human face</td> <td rowspan="2"><code>npm i @tensorflow-models/face-landmarks-detection</code></td> </tr> <tr> <td><a href="./face-landmarks-detection/demos">source</a></td> </tr>   <tr> <td rowspan="2"><b>Audio</b></td> <td rowspan="2"><b><a style="white-space:nowrap; display:inline-block;" href="./speech-commands"><div style='vertical-align:middle; display:inline;'>Speech Commands</div></a></b></td> <td><a href="https://storage.googleapis.com/tfjs-speech-model-test/2019-01-03a/dist/index.html">live</a></td> <td rowspan="2">Classify 1 second audio snippets from the <a href="https://www.tensorflow.org/tutorials/audio/simple_audio">speech commands dataset</a>.</td> <td rowspan="2"><code>npm i @tensorflow-models/speech-commands</code></td> </tr> <tr> <td><a href="./speech-commands/demo/index.html">source</a></td> </tr>   <tr> <td rowspan="4"><b>Text</b></td> <td rowspan="2"><b><a style="white-space:nowrap; display:inline-block;" href="./universal-sentence-encoder"><div style='vertical-align:middle; display:inline;'>Universal Sentence Encoder</div></a></b></td> <td><a href=""></a></td> <td rowspan="2">Encode text into a 512-dimensional embedding to be used as inputs to natural language processing tasks such as sentiment classification and textual similarity.</td> <td rowspan="2"><code>npm i @tensorflow-models/universal-sentence-encoder</code></td> </tr> <tr> <td><a href="./universal-sentence-encoder/demo">source</a></td> </tr>  <tr> <td rowspan="2"><b><a style="white-space:nowrap; display:inline-block;" href="./toxicity"><div style='vertical-align:middle; display:inline;'>Text Toxicity</div></a></b></td> <td><a href="https://storage.googleapis.com/tfjs-models/demos/toxicity/index.html">live</a></td> <td rowspan="2">Score the perceived impact a comment might have on a conversation, from "Very toxic" to "Very healthy".</td> <td rowspan="2"><code>npm i @tensorflow-models/toxicity</code></td> </tr> <tr> <td><a href="./toxicity/demo/index.html">source</a></td> </tr>   <tr> <td rowspan="2"><b>Depth Estimation</b></td> <td rowspan="2"><b><a style="white-space:nowrap; display:inline-block;" href="./depth-estimation"><div style='vertical-align:middle; display:inline;'>Portrait Depth</div></a></b></td> <td><a href="https://storage.googleapis.com/tfjs-models/demos/3dphoto/index.html">live</a></td> <td rowspan="2">Estimate per-pixel depth (the distance to the camera center) for a single portrait image, which can be further used for creative applications such as <a href="https://blog.tensorflow.org/2022/05/portrait-depth-api-turning-single-image.html?linkId=8063793">3D photo</a> and <a href="https://storage.googleapis.com/tfjs-models/demos/relighting/index.html">relighting</a>.</td> <td rowspan="2"><code>npm i @tensorflow-models/depth-estimation</code></td> </tr> <tr> <td><a href="./depth-estimation/demos/3d_photo/index.html">source</a></td> </tr>  <tr> <td rowspan="2"><b>General Utilities</b></td>  <td rowspan="2"><b><a style="white-space:nowrap; display:inline-block;" href="./knn-classifier"><div style='vertical-align:middle; display:inline;'>KNN Classifier</div></a></b></td> <td><a href=""></a></td> <td rowspan="2">This package provides a utility for creating a classifier using the K-Nearest Neighbors algorithm. Can be used for transfer learning.</td> <td rowspan="2"><code>npm i @tensorflow-models/knn-classifier</code></td> </tr> <tr> <td><a href="./knn-classifier/demo">source</a></td> </tr> </table>

Development

You can run the unit tests for any of the models by running the following inside a directory:

yarn test

New models should have a test NPM script (see this package.json and run_tests.ts helper for reference).

To run all of the tests, you can run the following command from the root of this repo:

yarn presubmit