JustDepth: Real-Time Radar-Camera Depth Estimation with Single-Scan LiDAR Supervision

2026-07-24 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, quick

Summary

JustDepth is a novel single-stage radar-camera depth estimator designed for autonomous systems, providing accurate, low-latency depth. It is trained using only radar, camera, and single-scan LiDAR data. The system aggregates all radar returns into a fixed-width 1D representation, ensuring runtime independence from point count. Key architectural components include a Height Fusion Block for modality integration, a lightweight Graph Neural Network (GNN) for global depth propagation, and a training-only confidence decoder that stabilizes learning without test-time overhead. JustDepth also employs simple augmentations to mitigate stripe artifacts, quantified by the Vertical-Horizontal Gradient Ratio (VHGR). On the nuScenes dataset, JustDepth achieves comparable accuracy to state-of-the-art methods while reducing inference time by 39.7x and stripe artifacts by 66%.

Key takeaway

For Computer Vision Engineers developing autonomous systems that demand real-time, accurate depth perception, JustDepth offers a significant advancement. Its single-stage radar-camera fusion architecture reduces inference time by 39.7x and minimizes stripe artifacts by 66% on nuScenes. You should evaluate this approach to streamline your perception pipelines, enhance system responsiveness, and potentially reduce computational overhead in your deployments.

Key insights

JustDepth integrates radar and camera for real-time depth estimation, using single-scan LiDAR for supervision.

Principles

• Decouple runtime from radar point count via fixed-width 1D representation.
• Use a training-only confidence decoder to stabilize learning.
• Mitigate stripe artifacts with simple augmentations.

Method

Aggregates radar returns into a fixed-width 1D representation, fuses modalities with a Height Fusion Block, and propagates depth globally via a lightweight GNN.

In practice

• Real-time depth for autonomous driving perception.
• Reduce inference latency in radar-camera systems.
• Improve depth accuracy in sparse radar environments.

Topics

• Radar-Camera Fusion
• Depth Estimation
• Autonomous Driving
• Real-Time Perception
• LiDAR Supervision
• Graph Neural Networks
• nuScenes Dataset

Best for: Research Scientist, AI Scientist, Computer Vision Engineer, Robotics Engineer

Related on AIssential

• JustDepth: Real-Time Radar-Camera Depth Estimation with Single-Scan LiDAR Supervision — Single-stage radar-camera depth estimation achieves real-time performance and artifact reduction with minimal supervision.
• ZipDepth: Bringing Lightweight Zero-Shot Monocular Depth Anywhere, on Any Device — ZipDepth enables robust zero-shot monocular depth on edge devices via efficient architecture and knowledge distillation.
• RCGDet3D: Rethinking 4D Radar-Camera Fusion-based 3D Object Detection with Enhanced Radar Feature Encoding — Optimizing 4D radar feature extraction significantly boosts 3D object detection accuracy and speed, outperforming complex fusion methods.
• Towards Robust Driving Perception: A Flexible Scale-Driven Family for Self-Supervised Monocular Depth Estimation — FlexDepth improves monocular depth estimation for driving by decoupling static/dynamic training and using a scale-driven decoder.
• Depth-Semantic Alignment and Affinity-Guided Fusion for Structured Radar Point Cloud Generation — Fusing vision and radar data improves sparse, noisy millimeter-wave radar point clouds for perception tasks.
• On Exploring Input Resolution Scaling For Anytime LiDAR Object Detection — Dynamically scaling LiDAR input resolution with a single DNN model enables efficient anytime 3D object detection for real-time systems.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.