IARPA WRIVA Research

My research is supported by the IARPA Walk-through Rendering from Images of Varying Altitude (WRIVA) program, which aims to develop software systems for site modeling in scenarios where only a limited amount of ground-level imagery with reliable metadata is available. My research focuses on unconstrained 3D reconstruction and novel view synthesis in challenging real-world environments, with contributions in the following areas:

• Efficient rendering, compression, and training for 3D Gaussian Splatting (3DGS) [1, 2, 3]
• Multi-view consistent super-resolution for training 3DGS with low-resolution imagery
• Image, video, and multiview diffusion model priors for sparse-view 3D reconstruction
• Uncertainty quantification algorithms for 3DGS and Neural Radiance Fields (NeRF)

The Principal Investigators overseeing my research include Professor Tom Goldstein, Professor Matthias Zwicker, Professor Abhinav Shrivastava, Dr. Abhay Yadav, Dr. Cheng Peng, and Professor Rama Chellappa. My team at the University of Maryland Institute of Advanced Computer Studies partners with Johns Hopkins University, with whom I previously collaborated on multimodal biometric recognition research under the IARPA BRIAR program at STR. Additional institutional partners include Rice University, the University of California San Diego, and Arizona State University.

Diffusion Model Priors for 3D Reconstruction

Diffusion in the Loop produces sharper reconstructions with fewer artifacts than the baseline 3D model in sparse-view reconstruction settings. Applying posthoc diffusion cleanup further enhances visual quality and fine detail.

3D/4D Research Highlights BibTeX References

	SpeeDe3DGS: Speedy Deformable 3D Gaussian Splatting with Temporal Pruning and Motion Grouping Accepted to CVPR, 2026. PAPER CODE WEBSITE Boost DeformableGS rendering speed from 20 to 276 FPS using temporal sensitivity pruning and groupwise SE(3) motion distillation, all while preserving the superior image quality of per-Gaussian neural motion. Allen Tu*, Haiyang Ying*, Alex Hanson, Yonghan Lee, Tom Goldstein, Matthias Zwicker
	SplatSuRe: Selective Super-Resolution for Multi-view Consistent 3D Gaussian Splatting Accepted to CVPR, 2026. PAPER CODE WEBSITE Enhance 3D Gaussian Splatting by selectively injecting super-resolution only where high-frequency detail is missing, yielding sharper results and improved perceptual quality without introducing multi-view inconsistencies. Pranav Asthana, Alex Hanson, Allen Tu, Tom Goldstein, Matthias Zwicker, Amitabh Varshney
	Speedy-Splat: Fast 3D Gaussian Splatting with Sparse Pixels and Sparse Primitives CVPR, 2025. PAPER CODE WEBSITE Accelerate 3D Gaussian Splatting rendering speed by 2× for free by accurately localizing primitives during rasterization and over 6× in total by pruning the scene by more than 90% during training, providing a significantly higher speedup than existing techniques while maintaining competitive image quality. Alex Hanson, Allen Tu, Geng Lin, Vasu Singla, Matthias Zwicker, Tom Goldstein
	PUP 3D-GS: Principled Uncertainty Pruning for 3D Gaussian Splatting CVPR, 2025. PAPER CODE WEBSITE Prune 90% of primitives from any pretrained 3D Gaussian Splatting model using a mathematically principled sensitivity score, more than tripling rendering speed while retaining more salient foreground information and higher visual fidelity than previous techniques at a substantially higher compression ratio. Alex Hanson*, Allen Tu*, Vasu Singla, Mayuka Jayawardhana, Matthias Zwicker, Tom Goldstein
	SPAR-3D: Security, Privacy, and Adversarial Robustness in 3D Generative Vision Models CVPR Workshop, 2026. WEBSITE CALL FOR PAPERS REVIEWERS The SPAR-3D workshop brings together the 3D vision, AI security, and multimodal reasoning communities to advance robustness, traceability, and trustworthy 3D generative systems. Organizers: Nicole Meng, Yingjie Lao, Francis Engelmann, Ethan Rathbun, Shaoyi Huang, Allen Tu, Josué Martínez-Martínez, Dongjin Song, Faysal Hossain Shezan, Renjie Wan

* denotes equal contribution.

Acknowledgements

This research is based upon work supported by the Office of the Director of National Intelligence (ODNI), Intelligence Advanced Research Projects Activity (IARPA), via IARPA R&D Contract No. 140D0423C0076. The views and conclusions contained herein are those of the authors and should not be interpreted as necessarily representing the official policies or endorsements, either expressed or implied, of the ODNI, IARPA, or the U.S. Government. The U.S. Government is authorized to reproduce and distribute reprints for Governmental purposes notwithstanding any copyright annotation thereon.