MonoDepth.zip CVPR 2026 Demo Proposal

Zero-Shot Real-Time Monocular Depth on Mobile Devices

Fabio Tosi1, Luca Bartolomei1, Matteo Poggi1, Stefano Mattoccia1
1University of Bologna

MonoDepth.zip estimates depth from a single image in real time, directly on device — no internet connection, no setup, no restrictions on scene type. All results shown on this page were captured live on an iPhone 12.

Overview

MonoDepth.zip is a monocular depth estimation system that runs in real time on commodity mobile hardware. Given a single RGB image, it produces a dense depth map covering the full field of view — instantly, on device.

The system generalizes across drastically different environments out of the box: indoor rooms, outdoor scenes, urban streets, and unstructured settings — all without any scene-specific configuration or retraining.

Our goal is to bring accurate, responsive depth perception to everyday devices, opening the door to real-world applications in augmented reality, robotics, accessibility, and computational photography.

Live on Device

The full pipeline runs on-device with no cloud dependency — point your iPhone at any scene and get depth instantly.

Works Everywhere, Out of the Box

One model. No fine-tuning. Across indoor, outdoor, driving, and beyond.

RGB vs. Depth — Driving

Drag the slider to reveal the RGB image or the predicted depth map.

RGB Depth

RGB vs. Depth — Nighttime

Robust depth estimation even under challenging low-light conditions.

RGB Depth

RGB vs. Depth — Outdoor

Drag the slider to reveal the RGB image or the predicted depth map.

RGB Depth