Skip to content

Latest commit

 

History

History

firebase-object-detection-node

README.md

TensorFlow.js Example: Running a TensorFlow SavedModel in Node.js

This demo demonstrates how to run a TensorFlow SavedModel in Node.js natively without using tfjs-converter to convert the model. Native execution of TensorFlow SavedModel in Node.js supports more ops and better performance compared with converted model.

This example runs a object detection model in Node.js, and hosts the model through Cloud Functions for Firebase. Before getting started, please complete the first three steps of the Firebase Guide. For more information about the Tensorflow object detection API, check out the README in tensorflow/object_detection.

To launch the demo, complete the following steps:

  1. Go to file .firebaserc and update the value projects.default with your own Firebase project ID.

  2. Run the following commands:

cd functions
npm install
firebase serve
  1. Open the website link displayed in the log. The link has the following format:

http://localhost:5001/[firebase-project-id]/us-central1/app

It will take several seconds to load and warm up the model. Once the page is loaded, you can upload an JPEG image and test. Following is an example output:

example output

By default, the inference uses tfjs-node, which runs on the CPU. If you have a CUDA-enabled GPU and have the CUDA and CuDNN libraries set up properly on your system, you can run the inference on the GPU by replacing the tfjs-node package with tfjs-node-gpu in the package.json file and functions/index.js file.

Please note: Node.js versions 10, 12, 14, and 16 are supported. See Set runtime options for important information regarding ongoing support for these versions of Node.js.