Skip to main content
AI RADARTools
423•08/19/2026, 22:15•2 min read

ML Project of the Week – Real-Time Scene Depth Estimation

Executive TL;DR30s read

A project by an engineer from Poland showcases the ability to estimate scene depth in real-time using a neural network. Utilizing the TypeGPU library, it achieves impressive results directly in the browser.

ML Project of the Week – Real-Time Scene Depth Estimation

A project by an engineer from Software Mansion (Poland) has garnered 7.4 million views in just one day. It works with TypeGPU – a library that allows writing GPU shaders directly in TypeScript.

The author launched a neural network that reconstructs a depth map of a scene from a regular 2D image, achieving this in about 8 ms on an M4 Pro chip.

A video frame at 60 fps lasts about 16.7 ms, making it a full real-time experience right in the browser. Additionally, everything from the neural network to the final render with lighting operates in a single GPU pipeline, without data transfer back and forth between the CPU and GPU.

The demo looks particularly impressive, but the code is not open-source (it might be sold to Zoom for several million).

Why it matters

Analysis

This project showcases new possibilities in real-time image processing, which could significantly impact the development of web technologies and AI. The use of GPU shaders in TypeScript opens new horizons for developers.

Discuss in community

Share your questions and insights with developers

Join Telegram
+5 Points
← Back to news feed

Related in Ecosystem

Practical guides, core terminology, and agent skills