Get Started
Home
Topics
Search
Library
Research questionHow can self-supervised monocular depth estimation stay accurate across scales and dynamic traffic on automotive edge devices?Monocular depth models trained without ground-truth depth often degrade in complex scenes containing moving vehicles and pedestrians, while models designed for such scenes can be too computationally expensive for edge deployment. Handling different scales without sacrificing depth precision further complicates the tradeoff.
AI
Computer Vision
Image & Video Processing
Machine Learning
Research Paper
Small / On-device Models
Technology
Latest papersRecent research connected to this question, newest first.Towards Robust Driving Perception: A Flexible Scale-Driven Family for Self-Supervised Monocular Depth EstimationThe source examines self-supervised monocular depth estimation for challenging driving scenarios and reports results across standard driving benchmarks. Its proposed model family uses static–dynamic decoupled training and scale-dependent decoding; the reported evidence includes operation without auxiliary information, zero-shot generalization, and a smallest model rated at 0.7 GFLOPs and 37.6 FPS on mobile platforms.research paper · Sep 4, 2026
Related questions
How can single-image depth estimation resolve scale ambiguity to produce consistent metric depth across environments?How can deep learning provide reliable depth estimates from monocular or stereo endoscopic images?How can vision-only BEV perception remain accurate while meeting real-time constraints on production hardware?How can monocular depth estimation remain accurate across adverse weather without harming clean-image performance?