Depthpro ONNX

🚀 DepthPro with Transformers.js for Depth Estimation

This project enables depth estimation using the apple/DepthPro model with ONNX weights, made compatible with the Transformers.js library.

🚀 Quick Start

This repository provides the apple/DepthPro model with ONNX weights, making it compatible with Transformers.js for depth estimation tasks.

📦 Installation

If you haven't already, you can install the Transformers.js JavaScript library from NPM using the following command:

npm i @huggingface/transformers

💻 Usage Examples

Basic Usage

import { AutoProcessor, AutoModelForDepthEstimation, RawImage } from "@huggingface/transformers";

// Load model and processor
const depth = await AutoModelForDepthEstimation.from_pretrained("onnx-community/DepthPro-ONNX", { dtype: "q4" });
const processor = await AutoProcessor.from_pretrained("onnx-community/DepthPro-ONNX");

// Read and prepare image
const image = await RawImage.read("https://raw.githubusercontent.com/huggingface/transformers.js-examples/main/depth-pro-node/assets/image.jpg");
const inputs = await processor(image);

// Run depth estimation model
const { predicted_depth, focallength_px } = await depth(inputs);

// Normalize the depth map to [0, 1]
const depth_map_data = predicted_depth.data;
let minDepth = Infinity;
let maxDepth = -Infinity;
for (let i = 0; i < depth_map_data.length; ++i) {
  minDepth = Math.min(minDepth, depth_map_data[i]);
  maxDepth = Math.max(maxDepth, depth_map_data[i]);
}
const depth_tensor = predicted_depth
  .sub_(minDepth)
  .div_(-(maxDepth - minDepth)) // Flip for visualization purposes
  .add_(1)
  .clamp_(0, 1)
  .mul_(255)
  .round_()
  .to("uint8");

// Save the depth map
const depth_image = RawImage.fromTensor(depth_tensor);
depth_image.save("depth.png");

The following images illustrate the input image and its corresponding depth map generated by the model:

Input Image	Depth Map

📄 License

This project is licensed under the apple-ascl license.

⚠️ Important Note

Having a separate repo for ONNX weights is intended to be a temporary solution until WebML gains more traction. If you would like to make your models web - ready, we recommend converting to ONNX using 🤗 Optimum and structuring your repo like this one (with ONNX weights located in a subfolder named onnx).