TensorFlow.js Interactive Model Visualizers

TensorFlow.js-powered Interactive Model Visualizers for standard perception tasks embeddable anywhere in the web.

Supported tasks:

Image Classification
Object Detection
Image Segmentation

The Interactive Visualizer supports any model coming with a metadata JSON file formatted following the supported tasks standards. This metadata file is passed at runtime to the visualizer as URL query parameter (e.g. https://visualizerHostedUrl/?modelMetadataUrl=https://myModelMetadataUrl/metadata.json).

The visualizer will then load its UI and behave according to the task defined by the model metadata.

NOTE: standards definition is WIP. Here's an example of such metadata for an image classifier model: https://storage.googleapis.com/tfhub-visualizers/google/aiy/vision/classifier/plants_V1/1/metadata.json.

This project was generated with Angular CLI version 10.1.3.

Development server

Run ng serve for a dev server. Navigate to http://localhost:4200/. The app will automatically reload if you change any of the source files.

Build

Run ng build to build the project. The build artifacts will be stored in the dist/ directory. Use the --prod flag for a production build.

Test

Run ng test to run the tests. Karma is providing the testing environment, and Jasmine is used for unit testing.

Additionally run ng lint to check for the linter warnings/errors.