retinanetjs

retinanetjs

This package provides some convenience methods for using TensorFlow models created using keras-retinanet.

Getting Started

Convert RetinaNet Model to TensorFlowJS

In Python, use the two functions defined in notebooks/Functions.ipynb to convert an existing RetinaNet model to a RetinaNetJS-compatible TensorFlowJS folder. The only dependencies are keras_retinanet and tensorflowjs.

The two functions (save_retinanet_to_savedmodel and convert_retinanet_savedmodel_to_tfjs) have to be used in two separate Python sessions (e.g., in Jupyter, you must restart the kernel).

For example, to convert the pretrained ResNet50 model, you can follow the example in notebooks/Convert COCO (ResNet50).ipynb.

Using Model with retinanetjs

This is the (easier) part. See examples for use cases.

import { load } from 'retinanetjs'

detector = await load('http://www.example.com/path/to/coco_resnet50', COCO_CLASSES)