SADGE: Structure and Appearance Domain Gap Estimation of Synthetic and Real Data

2026-05-21 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision · Depth: Expert, quick

Summary

SADGE is a quantitative similarity metric designed to predict the performance of synthetic image datasets for common computer vision tasks without requiring downstream model training. This addresses a key bottleneck in model development, as existing metrics like PSNR, FID, and CLIP primarily measure semantic alignment (Appearance Similarity), while structural similarity (Geometric Similarity) is less commonly considered. Research shows that neither appearance nor geometry alone reliably predicts downstream performance; instead, their non-linear interplay dictates synthetic data utility. SADGE measures the correlation between these metrics and downstream performance in object detection, semantic segmentation, and pose estimation. Across five public synthetic-to-real benchmark families and 15 dataset-level variants (79k image pairs), SADGE achieved Pearson r=0.88 and Spearman rho=0.77. The optimal configuration fuses DINOv3 appearance similarity with MASt3R geometric consistency through a constrained bilinear interaction.

Key takeaway

For Computer Vision Engineers evaluating synthetic datasets, relying solely on appearance or geometric similarity metrics is insufficient. Your synthetic data utility is best predicted by the non-linear interaction of both factors. Consider SADGE's approach, which fuses DINOv3 appearance with MASt3R geometry. This can accurately forecast downstream performance in tasks like object detection and semantic segmentation, saving significant training time.

Key insights

Synthetic data utility for computer vision tasks is best predicted by the non-linear interplay of appearance and geometric similarities, not either alone.

Principles

• Neither appearance nor geometry alone predicts synthetic data utility.
• Non-linear interplay of appearance and geometry dictates synthetic data utility.
• Fusing DINOv3 and MASt3R provides optimal similarity prediction.

Method

SADGE estimates synthetic data utility by fusing DINOv3 appearance similarity with MASt3R geometric consistency through a constrained bilinear interaction, correlating with downstream performance.

In practice

• Evaluate synthetic datasets without model training.
• Improve object detection with better synthetic data.
• Enhance semantic segmentation performance.

Topics

• SADGE
• Synthetic Data Evaluation
• Computer Vision
• Domain Gap Estimation
• Object Detection
• Semantic Segmentation

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Computer Vision Engineer

Related on AIssential

• Training-Free Metrics for Synthetic Object Detection Data: A Proxy for Detector Performance — CCDM metrics provide a training-free, highly accurate proxy for evaluating synthetic object detection data utility, outperforming existing methods.
• Towards Consistent Video Geometry Estimation — ViGeo uses dynamic chunking attention and data refinement for consistent video geometry estimation across various inference modes.
• Synthetic Image Detection with CLIP: Understanding and Assessing Predictive Cues — CLIP-based synthetic image detectors rely on high-level photographic attributes, not just artifacts, but struggle with cross-generator generalization.
• H-OmniStereo: Zero-Shot Omnidirectional Stereo Matching with Heading-Aligned Normal Priors — H-OmniStereo improves omnidirectional stereo matching via a large synthetic dataset and a heading-aligned normal estimator.
• Training-free Cross-domain Few-shot Segmentation via Robust Semantic Representation and Matching — A training-free framework built on DINOv3 achieves state-of-the-art CD-FSS by enhancing semantic representation and matching without overfitting.
• Beyond Real Data: Synthetic Data through the Lens of Regularization — An optimal synthetic-to-real data ratio exists, minimizing generalization error based on distribution similarity.

Open in AIssential →

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