SPAR-3D: Security, Privacy, and Adversarial Robustness in 3D Generative Vision Models

With rapid advances in computer vision and graphics, 3D scene reconstruction and understanding have become increasingly central to modern visual computing. Methods such as Neural Radiance Fields (NeRF), 3D Gaussian Splatting, 3D diffusion, and vision-language-3D frameworks enable real-time scene rendering and are widely applied in drone operations, cultural heritage, and autonomous driving. However, these systems face security, privacy, and provenance challenges, including adversarial geometry perturbations, data leakage, and model inversion. The SPAR-3D workshop brings together the 3D vision, AI security, and multimodal reasoning communities to advance robustness, traceability, and trustworthy 3D generative systems.

Call For Papers

We invite submissions in, but not limited to, the following areas:

Threats and Vulnerabilities in 3D AI
• Robustness and adversarial attack and defense for 3D generative models such as NeRF based methods, 3D Gaussian Splatting, 3D diffusion models, point based representations, 3D aware GANs, and scene graph generative models
• Security and robustness of 3D perception and reasoning models such as point cloud networks, 3D transformers, 3D vision language models, 3D semantic segmentation, 3D object detection, scene graph learning, and large world models
• Data poisoning and multi view consistency attacks on 3D reconstruction and novel view synthesis models
• Privacy leakage, inversion, attribute inference, and provenance tracking in 3D datasets and neural scene representations

Content Authentication, Ownership, and Provenance
• Scene fingerprinting, spatial watermarking, steganography, and ownership verification in 3D assets and neural fields
• Provenance analysis, dataset lineage tracking, and manipulation tracing in 3D content generation pipelines

Secure Generation, Defense, and Evaluation
• Secure and trustworthy 3D content generation, manipulation, and editing using text prompts, diffusion models, image conditioning, masks, or interactive interfaces
• Defensive methods against security and privacy threats in 3D generative, reconstructive, and perception models
• Red teaming, evaluation, and benchmarking for generative and reconstructive 3D and 4D pipelines
• 3D and 4D datasets, simulations, benchmarks, and metrics for spatial robustness and trustworthiness
• Standardized open evaluation, reproducible pipelines, and policy implications for 3D generative system safety

Multimodality, Applications, and Societal Impact
• Secure multimodal reasoning in vision language 3D models, world simulators, robotics, and embodied agents
• Applications of secure 3D AI in robotics, autonomous driving, AR and VR, digital twins, and embodied perception
• Ethical, societal, and regulatory frameworks for secure and responsible deployment of 3D AI

Submissions are full length papers (up to 8 pages), including figures and tables, in the CVPR style. Additional pages containing only cited references are allowed. Please download the CVPR 2026 Submission Template for detailed formatting instructions.

Among the accepted papers, two will be selected for oral presentations, while the remaining papers will be presented as posters. All submissions will be managed through OpenReview.