https://github.com/ely-s/efficientdetjs

Port of the EfficientDet neural network architecture to the web.
https://github.com/ely-s/efficientdetjs

ai cv tensorflow

Last synced: about 1 year ago
JSON representation

Port of the EfficientDet neural network architecture to the web.

• Host: GitHub
• URL: https://github.com/ely-s/efficientdetjs
• Owner: Ely-S
• Created: 2020-01-03T20:52:11.000Z (over 6 years ago)
• Default Branch: master
• Last Pushed: 2023-01-06T16:11:25.000Z (over 3 years ago)
• Last Synced: 2025-04-13T14:13:11.415Z (about 1 year ago)
• Topics: ai, cv, tensorflow
• Language: TypeScript
• Homepage: https://ondaka.github.io/EfficientDetJS/index.html
• Size: 123 MB
• Stars: 4
• Watchers: 1
• Forks: 2
• Open Issues: 26
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
### EfficientDetJs

This library lets you easily deploy a State-of-the-Art EfficientDet model to a tensorflow.js runtime. EfficientDet is a fast, very powerful neural architecture with an active open source [implementation](https://github.com/google/automl) making it a good base for new projects. This library allows you to use a pretrained or custom EfficientDet model without messing with tensorflow.js yourself.

[See a live demo](https://ondaka.github.io/EfficientDetJS/example/dist/).

## Getting Started

    import EfficientDet from "EfficientDetJS"

    

    const model = new EfficientDet()

    // load the model from tf hub

    await model.load()

    // get an array of object bounding boxes

    // .predict accepts a a Tensor3D of an image.

    const predictions = model.predict(image)

    // draw boxes on canvas

    model.draw(predictions, document.getElementById("mycanvas))

### Pretrained checkpoint

The pretraiend checkpoint, efficientdet-d0 is trained on a 90 class COCO challange. It is hosted here on tensorflow hub

### Custom checkpoints

For custom efficientdet models, refer to the Dockerfile for details on how to export your own model from EfficientDet

###  Building the typescript

    yarn build

#### Building the model

There is a Dockerfile in hub/ that will build an image containing an EfficientDet model and convert it to the tensorflow.js format

    cd hub

    docker build -t efficientdet-model-d0 .

    

    # Or to build a different moodel size

    # docker build --build-arg SIZE=d1 -t efficientdet-model-d0 .

    # Copy exported model files into current directory

    docker run -v (pwd):/out efficientdet-model-d0  cp -r /tmp/efficientdet-d0.js /out/