PriSAR: 3D Geometric-Prior-Guided Diffusion for Parameter-Controlled SAR Image Generation

2026-07-24

AI Summary

This work addresses the challenges of controllable synthetic aperture radar (SAR) image generation under sparse observation angles and the lack of explicit geometric guidance by proposing a diffusion-based generative method that incorporates 3D geometric priors. The approach leverages a lightweight multi-bounce ray tracing algorithm to extract geometric priors, which are then integrated with point cloud encoding and text conditioning into the Stable Diffusion 3.5 Medium architecture. Efficient fine-tuning is achieved via low-rank adaptation (LoRA). The proposed method substantially improves viewpoint consistency and geometric fidelity of generated SAR images, achieving an SSIM of 0.812 and azimuth consistency of 0.940 on four real aircraft datasets, and an SSIM of 0.878 with azimuth consistency of 0.917 on five vehicle classes from the MSTAR dataset.

Abstract

Synthetic aperture radar (SAR) image generation can mitigate data scarcity, but controllablegeneration under sparse observation angles remains difficult. Recent SAR generative studies im-prove texture realism, yet explicit geometry-aware control is still limited. This paper studiesthe focused and verifiable setting of intermediate-azimuth completion: 3D-model-derived geo-metric priors guide a diffusion model to synthesize the views missing from sparse-angle trainingdata. GeoDiff-SAR constructs a lightweight multi-bounce ray-tracing prior, encodes the result-ing point cloud, and fuses it with text conditioning while adapting Stable Diffusion 3.5 Mediumthrough low-rank adaptation. On a real four-category aircraft dataset, GeoDiff-SAR reaches anSSIM of 0.812 and azimuth consistency of 0.940, compared with 0.738 and 0.782 for the text-conditioned SD3.5 Medium baseline. The same sparse-angle protocol on five MSTAR vehicleclasses yields an SSIM of 0.878 and azimuth consistency of 0.917. These results support theconclusion that a lightweight 3D geometric prior improves viewpoint adherence for controllableSAR generation; it is intended as generation guidance rather than high-fidelity electromagneticreconstruction.

Problem

• SAR image generation
• controllable generation
• sparse observation angles
• geometry-aware control
• data scarcity

Innovation

• 3D geometric prior
• diffusion model
• SAR image generation
• low-rank adaptation
• azimuth consistency