Publications

Vibhas Kumar Vats, David Crandall, Samuel Goree • ArXiv-preprint

This paper investigates why generative models can degrade when they are repeatedly trained on AI-generated data. We introduce the idea of neural resonance, where iterative feedback pushes models toward a low-dimensional latent structure, and show the conditions under which this happens. We also analyze several model families and describe common patterns of collapse that can help guide future mitigation strategies.

DiffusionGenerative ModelsDeep LearningNeural Resonance

arXiv BibTeX

Vibhas Kumar Vats, David Crandall • ACM Computing Surveys

This survey examines how geometry can be integrated into deep learning frameworks for depth estimation and related computer vision problems. It presents a taxonomy of common geometry-based constraints, compares existing methods, and discusses future research opportunities.

3D ReconstructionDepth EstimationMulti-View StereoSurvey

PDF arXiv BibTeX DOI

Huyung Ha, Vibhas Kumar Vats, Soon-heung Jung, Alimoor Reza, David Crandall • ICCV

This paper presents an efficient method for point-cloud upsampling from sparse or incomplete 3D data. We propose Hybrid Voxels, which combine the efficiency of voxel-based processing with the precision of point-based representations, and build HVPUNet on top of this design. The resulting framework restores missing geometry, refines surface detail, and achieves strong accuracy with lower computational cost.

3D ReconstructionSuper ResolutionPoint Cloud Upsampling

PDF BibTeX

Vibhas Vats, Sripad Joshi, David Crandall, Md. Alimoor Reza, Soon-heung Jung • WACV

GC-MVSnet explicitly models geomtric consistency constraints across different views to acceralte optimization and geometric understanding of learning-based models. We produce state-of-the-art results on DTU, BlendedMVS, and Tanks & Temples datasets.

3D ReconstructionMulti-View Stereo

Project PDF arXiv Code BibTeX Video Supplement Journal Version

Vibhas Kumar Vats, David Crandall, Samuel Goree • ArXiv-preprint

This paper investigates why generative models can degrade when they are repeatedly trained on AI-generated data. We introduce the idea of neural resonance, where iterative feedback pushes models toward a low-dimensional latent structure, and show the conditions under which this happens. We also analyze several model families and describe common patterns of collapse that can help guide future mitigation strategies.

DiffusionGenerative ModelsDeep LearningNeural Resonance

arXiv BibTeX

Vibhas Vats, Md. Alimoor Reza, David Crandall, Soon-heung Jung • Neurocomputing

GC-MVSNet++ improves the reconstruction quality of 3D scenes by introducing a novel cost-regularization network based on the principles of dense-connnections. It produces competitive results to state-of-the-art MVSFormer++ paper without using Transformers.

3D ReconstructionMulti-View Stereo

arXiv Code BibTeX DOI Paper

Vibhas Kumar Vats, David Crandall • ACM Computing Surveys

This survey examines how geometry can be integrated into deep learning frameworks for depth estimation and related computer vision problems. It presents a taxonomy of common geometry-based constraints, compares existing methods, and discusses future research opportunities.

3D ReconstructionDepth EstimationMulti-View StereoSurvey

PDF arXiv BibTeX DOI

Vibhas Kumar Vats, Debvrat Varshney, Bhartendu Pandey, Christa Brelsford, Philipe Dias • SIGSPATIAL

This paper presents a generative AI approach for forecasting land-use and land-cover change over time. Using a diffusion model conditioned on historical and auxiliary data, we predict future imperviousness patterns across the United States at decadal scale. Our results show that generative models can provide useful large-scale forecasts and open new directions for physically informed Earth system prediction.

Generative ModelsDiffusionGeoAILulc

PDF BibTeX DOI

Huyung Ha, Vibhas Kumar Vats, Soon-heung Jung, Alimoor Reza, David Crandall • ICCV

This paper presents an efficient method for point-cloud upsampling from sparse or incomplete 3D data. We propose Hybrid Voxels, which combine the efficiency of voxel-based processing with the precision of point-based representations, and build HVPUNet on top of this design. The resulting framework restores missing geometry, refines surface detail, and achieves strong accuracy with lower computational cost.

3D ReconstructionSuper ResolutionPoint Cloud Upsampling

PDF BibTeX

Vibhas Kumar Vats, Zachary Wilkerson, Hiroki Sato, David Leake, David Crandall • ICCBR

This paper explores feature learning for case-based reasoning using deep neural networks. We introduce a training strategy that combines standard classification loss with additional objectives that better reflect how case-based systems compare examples. Experiments on multiple image datasets show that the learned features lead to stronger case-based classification than conventional end-to-end deep models.

CbrCase Based Reasoning

PDF BibTeX

Chuhua Wang, Md. Alimoor Reza, Vibhas Kumar Vats, Yingnan Ju, Nikhil Thakurdesai, Yuchen Wang, David Crandall, Soon-heung Jung, Jeongil Seo • Neurocomputing

This survey presents an overview of deep learning approaches for 3D reconstruction. It groups existing work into major problem settings and compares methods based on their architecture, outputs, datasets, and quantitative results. The paper also highlights current challenges and promising directions for future research.

Survey3D ReconstructionMulti-View Stereo

BibTeX DOI Paper

Zachary Wilkerson, David Leake, Vibhas Vats, David Crandall • ICCBR

This paper examines whether transfer learning can help learn better indexing features for case-based reasoning in limited-data domains. Our experiments show that pretrained deep models produce stronger and more reliable features for case retrieval than models trained from scratch. In the tested setting, the resulting case-based classifier also performs better than similar deep learning baselines.

CbrCase Based Reasoning

PDF BibTeX

Vibhas Vats, Sripad Joshi, David Crandall, Md. Alimoor Reza, Soon-heung Jung • WACV

GC-MVSnet explicitly models geomtric consistency constraints across different views to acceralte optimization and geometric understanding of learning-based models. We produce state-of-the-art results on DTU, BlendedMVS, and Tanks & Temples datasets.

3D ReconstructionMulti-View Stereo

Project PDF arXiv Code BibTeX Video Supplement Journal Version

David Leake, Zachary Wilkerson, Vibhas Vats, Karan Acharya, David Crandall • ICCBR

This paper examines how different deep learning architectures influence feature extraction for case-based classification. By comparing multiple models, we show that the usefulness of extracted features depends on both the network design and the data available for training. The study also provides guidance for choosing suitable models for case-based reasoning tasks.

Case Based ReasoningCbr

PDF BibTeX

Vibhas Kumar Vats, David Crandall • AAAI - workshop

This paper examines why some teacher models transfer knowledge more effectively than others in knowledge distillation. We find that the teacher’s training process strongly influences how much useful class relationship information is available for the student to learn. The study provides practical guidance for training teacher models to achieve better distillation results.

Deep LearningDistillation

arXiv BibTeX

Vibhas Kumar Vats, Sneha Rai, Mala De, Suddhasil De • Nanoelectronics, Circuits and Communication Systems

This paper addresses day-ahead load forecasting in situations where smart meter data is unavailable due to communication link failures. It proposes a classification-based framework that clusters loads by consumption level and uses those groups to improve forecasting under missing-data conditions. Experiments on real campus smart meter data demonstrate the practical value of the approach.

Power Systems

PDF BibTeX DOI

Vibhas Kumar Vats, Sneha Rai, Dibya Bharti, Mala De • ICCCA

This paper develops a load forecasting approach for smart meter data that works across several forecasting ranges, from 15-minute and hour-ahead prediction to day-ahead, month-ahead, and season-ahead forecasting. We evaluate the method on a real campus distribution system with mixed load types, showing its usefulness in practical and diverse energy settings.

Power Systems

PDF BibTeX DOI