Semi-SD: Semi-Supervised Metric Depth Estimation via Surrounding Cameras for Autonomous Driving

arXiv 25.3 2025 49.7 method

TLDR

Semi-SMD proposes a semi-supervised metric depth estimation framework using surrounding cameras and spatial-temporal-semantic fusion, achieving state-of-the-art on DDAD and nuScenes.

Reasoning

The paper introduces a novel fusion module and pose estimation method to address scale ambiguity in multi-camera setups, with strong empirical results. However, the integration of world models is used only as supervision and not as a core contribution, limiting novelty in that aspect.

Read-first score

Read-first score 49.7, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 2.

Recency 8%

86.7

Uses a gentle age decay so recent papers surface without erasing older foundations. 2025

Reproducibility 25%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=True; dataset=False; markers=code,dataset,github

Methodology quality 25%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=dataset,evaluation,metric,result

Topical relevance 42%

2.9

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Field roles

FrontierMethodology anchorReproducibility anchor

Rank sensitivity

Stability: volatile; rank range: 762.

Keyword Scores

world model

world simulator

generative world model

interactive world model

video world model

world dynamics prediction

model-based reinforcement learning world model

Deep Analysis

Innovations

Unified spatial-temporal-semantic fusion module to construct visual fused features
Cross-attention components for surrounding cameras and adjacent frames for metric scale refinement and temporal feature matching
Pose estimation framework using surrounding cameras, estimated depths, and extrinsic parameters to address scale ambiguity in multi-camera setups
Integration of semantic world model and monocular depth estimation world model to supervise depth estimation

Methodology

The framework takes adjacent surrounding frames and camera parameters as input. It uses a unified spatial-temporal-semantic fusion module with cross-attention to refine metric scale and match temporal features, and a pose estimation framework leveraging surrounding cameras, estimated depths, and extrinsic parameters to resolve scale ambiguity. Semantic and monocular depth world models provide supervision. Evaluation is performed on DDAD and nuScenes datasets.

Key Results

The method achieves state-of-the-art performance in surrounding camera based depth estimation quality on the DDAD and nuScenes datasets.

Resources

Paper Code