Depth Estimation

This package provides models for running depth estimation in TensorFlow.js.

Currently, we provide 1 model option:

AR Portrait Depth API

This AR portrait depth model estimates per-pixel depth (the distance to the camera center) for a single portrait image, which can be further used for creative applications. (See DepthLab for potential applications).

For example, the following demo transforms a single 2D RGB image into a 3D Portrait: 3D Photo Demo

Table of Contents

• How to Run It
• Example Code and Demos

How to Run It

There are two steps to run the AR portrait depth API:

First, you create an estimator by choosing one of the models from SupportedModels.

For example:

const model = depthEstimation.SupportedModels.ARPortraitDepth;
const estimator = await depthEstimation.createEstimator(model);

Next, you can use the estimator to estimate depth.

const estimationConfig = {
  minDepth: 0,
  maxDepth: 1,
}
const depthMap = await estimator.estimateDepth(image, estimationConfig);

The returned depth map contains depth values for each pixel in the image.

Example output:

{
  toCanvasImageSource(): ...
  toArray(): ...
  toTensor(): ...
  getUnderlyingType(): ...
}

The output provides access to the underlying depth values using the conversion functions toCanvasImageSource, toArray, and toTensor depending on the desired output type. Note that getUnderlyingType can be queried to determine what is the type being used underneath the hood to avoid expensive conversions (such as from tensor to image data).

Refer to each model's documentation for specific configurations for the model and their performance.

ARPortraitDepth Documentation

Example Code and Demos

You may reference the demos for code examples. Details for how to run the demos are included in the demos/ folder.