Alessio Quercia
Alessio Quercia
Home
Experience
Publications
Talks
Awards
Contact
1
SciFusion: Physics-Guided Compositional Diffusion for Multimodal 3D Scientific Image Generation
We present SciFusion, a unified 3D latent diffusion framework that combines compositional conditioning and domain-specific physics constraints to improve cross-domain scientific image generation and enable zero-shot synthesis across 15 imaging modalities.
Louxi Wang
,
Alessio Quercia
,
Hanno Scharr
,
Arya Bangun
PDF
Cite
3D-DOPAMIN: Diffusion-Optimized Projection Algorithm for 3D Multi-Coil MRI Reconstruction
We propose 3D-DOPAMIN, a hybrid framework that integrates physics-based constraints with a 3D diffusion model to achieve robust, high-fidelity 3D multi-coil MRI reconstruction from accelerated, undersampled data.
Alessio Quercia
,
Arya Bangun
,
Ira Assent
,
Hanno Scharr
PDF
Cite
Least but not Last: Fine-tuning Intermediate Principal Components for Better Performance-Forgetting Trade-Offs
This paper presents a comprehensive analysis of the performance-forgetting trade-offs inherent in low-rank adaptation using principal components as initialization.
Alessio Quercia
,
Arya Bangun
,
Ira Assent
,
Hanno Scharr
PDF
Cite
Physics-Guided Diffusion Priors for Multi-Slice Reconstruction in Scientific Imaging
We propose a memory-efficient framework combining partitioned diffusion priors with physics-based constraints to outperform existing multi-slice imaging baselines in accuracy and generalization across modalities like MRI and 4D-STEM.
Laurentius Valdy
,
Richard D. Paul
,
Alessio Quercia
,
Zhuo Cao
,
Xuan Zhao
,
Hanno Scharr
,
Arya Bangun
PDF
Cite
1LoRA: Summation Compression for Very Low-Rank Adaptation
We propose 1LoRA (Summation Low-Rank Adaptation), a compute, parameter and memory efficient fine-tuning method which uses the feature sum as fixed compression and a single trainable vector as decompression.
Alessio Quercia
,
Zhuo Cao
,
Arya Bangun
,
Richard D. Paul
,
Abigail Morrison
,
Ira Assent
,
Hanno Scharr
PDF
Cite
Code
Self-Supervised Learning based on Transformed Image Reconstruction for Equivariance-Coherent Feature Representation
We propose a self-supervised learning approach where the system learns transformations independently by reconstructing images that have undergone previously unseen transformations.
Qin Wang
,
Alessio Quercia
,
Benjamin Bruns
,
Abigail Morrison
,
Hanno Scharr
,
Kai Krajsek
PDF
Cite
Enhancing Monocular Depth Estimation with Multi-Source Auxiliary Tasks
We propose to improve MDE while using auxiliary datasets from related vision tasks with an alternating training scheme with a shared decoder built on top of a pre-trained vision foundation model.
Alessio Quercia
,
Erenus Yildiz
,
Zhuo Cao
,
Kai Krajsek
,
Abigail Morrison
,
Ira Assent
,
Hanno Scharr
PDF
Cite
Code
MRI Reconstruction with Regularized 3D Diffusion Model (R3DM)
We propose a 3D MRI reconstruction method that leverages a regularized 3D diffusion model combined with optimization method.
Arya Bangun
*
,
Zhuo Cao
*
,
Alessio Quercia
*
,
Hanno Scharr
,
Elizabeth Pfaehler
PDF
Cite
Code
Parameter-efficient Bayesian Neural Networks for Uncertainty-aware Depth Estimation
In this work, we investigate the suitability of PEFT methods for subspace Bayesian inference in large-scale Transformer-based vision models.
Richard D. Paul
*
,
Alessio Quercia
*
,
Vincent Fortuin
,
Katharina Nöh
,
Hanno Scharr
PDF
Cite
SGD Biased towards Early Important Samples for Efficient Training
Biasing SGD towards samples found to be more important early in training to improve training efficiency.
Alessio Quercia
,
Abigail Morrison
,
Hanno Scharr
,
Ira Assent
PDF
Cite
Code
»
Cite
×