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.

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.

This paper presents a comprehensive analysis of the performance-forgetting trade-offs inherent in low-rank adaptation using principal components as initialization.

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.

We propose a self-supervised learning approach where the system learns transformations independently by reconstructing images that have undergone previously unseen transformations.

We propose Focal Sampling, a method that biases SGD towards samples that are found to be more important after a few training epochs, by sampling them more often for the rest of the training.

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.

We propose a 3D MRI reconstruction method that leverages a regularized 3D diffusion model combined with optimization method.