publications

2026

1. MIDL 2026

   

   PIKACHU: Prototypical In-context Knowledge Adaptation for Clinical Heterogeneous Usage

   Amar Kumar, Zahra TehraniNasab, Emily Kaczmarek, and 1 more author

   In Medical Imaging with Deep Learning, 2026

   Abs Website

   Medical imaging systems increasingly rely on large vision language foundation models (VLFMs) trained on diverse biomedical corpora, yet these models remain difficult to adapt to new clinical tasks without costly fine-tuning and large annotated datasets. We present PIKACHU (Prototypical In-Context Knowledge Adaptation for Clinical Heterogeneous Usage), a lightweight and generalizable framework that enables rapid few-shot adaptation of frozen medical FMs using only a handful of labeled examples. Unlike prior approaches that modify backbone weights or introduce heavy attention-based adapters, PIKACHU performs all task adaptation directly in the FM feature space through in-context prototypical reasoning. Given a small support set, the framework constructs class prototypes by averaging normalized embeddings from a frozen VLFM image encoder and performs prediction on query images using temperature-scaled cosine similarity. Only a single temperature parameter is learned. We evaluate PIKACHU across three heterogeneous medical imaging datasets - dermatological images (ISIC), Optical Coherence Tomography (OCT), and Diabetic Retinopathy (DR), using established vision models (SigLIP, PubMedCLIP, DinoV2, and ViT) as backbones. The proposed in-context learning (ICL) strategy consistently outperforms the baseline (zero-shot) approaches across all datasets and architectures, achieving substantial improvements in both accuracy and AUC. Notably, with PubMedCLIP as the backbone, PIKACHU achieves 0.69/0.76 (Acc./AUC) on the ISIC dataset, 0.72/0.78 on OCT, and 0.79/0.88 on DR, demonstrating robust generalization across diverse clinical imaging modalities. These results highlight the promise of feature-space in-context learning as efficient and deployable paradigms for test-time adaptation of foundation models, without the need for extensive retraining.

2025

1. NeurIPS 2025

   

   Discovering Latent Graphs with GFlowNets for Diverse Conditional Image Generation

   Bailey Trang, Parham Saremi, Alan Q. Wang, and 6 more authors

   In The Thirty-ninth Annual Conference on Neural Information Processing Systems (NeurIPS), 2025

   Abs Website

   Capturing diversity is crucial in conditional and prompt-based image generation, particularly when conditions contain uncertainty that can lead to multiple plausible outputs. To generate diverse images reflecting this diversity, traditional methods often modify random seeds, making it difficult to discern meaningful differences between samples, or diversify the input prompt, which is limited in verbally interpretable diversity. We propose \modelnamenospace, a novel conditional image generation framework, applicable to any pretrained conditional generative model, that addresses inherent condition/prompt uncertainty and generates diverse plausible images. \modelname is based on a simple yet effective idea: decomposing the input condition into diverse latent representations, each capturing an aspect of the uncertainty and generating a distinct image. First, we integrate a latent graph, parameterized by Generative Flow Networks (GFlowNets), into the prompt representation computation. Second, leveraging GFlowNets’ advanced graph sampling capabilities to capture uncertainty and output diverse trajectories over the graph, we produce multiple trajectories that collectively represent the input condition, leading to diverse condition representations and corresponding output images. Evaluations on natural image and medical image datasets demonstrate \modelnamenospace’s improvement in both diversity and fidelity across image synthesis, image generation, and counterfactual generation tasks.

2. CVPR 2025 - MIV

   

   Language-Guided Trajectory Traversal in Disentangled Stable Diffusion Latent Space for Factorized Medical Image Generation

   Zahra TehraniNasab, Amar Kumar, and Tal Arbel

   Mechanistic Interpretability Workshop (MIV) [Proceeding Track] held in conjunction with IEEE / CVF Computer Vision and Pattern Recognition Conference (CVPR), Nashville, USA, June 10-17, 2025, 2025

   Abs arXiv Website

   Text-to-image diffusion models have demonstrated a remarkable ability to generate photorealistic images from natural language prompts. These high-resolution, language-guided synthesized images are essential for the explainability of disease or exploring causal relationships. However, their potential for disentangling and controlling latent factors of variation in specialized domains like medical imaging remains under-explored. In this work, we present the first investigation of the power of pre-trained vision-language foundation models, once fine-tuned on medical image datasets, to perform latent disentanglement for factorized medical image generation and interpolation. Through extensive experiments on chest X-ray and skin datasets, we illustrate that fine-tuned, language-guided Stable Diffusion inherently learns to factorize key attributes for image generation, such as the patient’s anatomical structures or disease diagnostic features. We devise a framework to identify, isolate, and manipulate key attributes through latent space trajectory traversal of generative models, facilitating precise control over medical image synthesis.

3. MICCAI 2025 - DGM4MICCAI

   

   Pixel Perfect MegaMed: A Megapixel-Scale Vision-Language Foundation Model for Generating High Resolution Medical Images

   Zahra TehraniNasab, Hujun Ni, Amar Kumar, and 1 more author

   In MICCAI Workshop on Deep Generative Models, 2025

   Abs arXiv Website

   Medical image synthesis presents unique challenges due to the inherent complexity and high-resolution details required in clinical contexts. Traditional generative architectures such as Generative Adversarial Networks (GANs) or Variational Auto Encoder (VAEs) have shown great promise for high-resolution image generation but struggle with preserving fine-grained details that are key for accurate diagnosis. To address this issue, we introduce Pixel Perfect MegaMed, the first vision-language foundation model to synthesize images at resolutions of 1024 × 1024. Our method deploys a multi-scale transformer architecture designed specifically for ultra-high resolution medical image generation, enabling the preservation of both global anatomical context and local image-level details. By leveraging vision-language alignment techniques tailored to medical terminology and imaging modalities, Pixel Perfect MegaMed bridges the gap between textual descriptions and visual representations at unprecedented resolution levels. We apply our model to the CheXpert dataset and demonstrate its ability to generate clinically faithful chest X-rays from text prompts. Beyond visual quality, these high-resolution synthetic images prove valuable for downstream tasks such as classification, showing measur- able performance gains when used for data augmentation, particularly in low-data regimes.

4. MICCAI 2025 - ELAMI

   

   Pixels Under Pressure: Exploring Fine-Tuning Paradigms for Foundation Models in High-Resolution Medical Imaging

   Zahra TehraniNasab, Amar Kumar, and Tal Arbel

   In International Workshop on Emerging LLM/LMM Applications in Medical Imaging, 2025

   Abs arXiv

   Advancements in diffusion-based foundation models have improved text-to-image generation, yet most efforts have been limited to low-resolution settings. As high-resolution image synthesis becomes increasingly essential for various applications, particularly in medical imaging domains, fine-tuning emerges as a crucial mechanism for adapting these powerful pre-trained models to task-specific requirements and data distributions. In this work, we present a systematic study, examining the impact of various fine-tuning techniques on image generation quality when scaling to high resolution 512x512 pixels. We benchmark a diverse set of fine-tuning methods, including full fine-tuning strategies and parameter-efficient fine-tuning (PEFT). We dissect how different fine-tuning methods influence key quality metrics, including Fréchet Inception Distance (FID), Vendi score, and prompt-image alignment. We also evaluate the utility of generated images in a downstream classification task under data-scarce conditions, demonstrating that specific fine-tuning strategies improve both generation fidelity and downstream performance when synthetic images are used for classifier training and evaluation on real images.

5. MICCAI 2025

   

   RL4Med-DDPO: reinforcement learning for controlled guidance towards diverse medical image generation using vision-language foundation models

   Parham Saremi, Amar Kumar, Mohamed Mohamed, and 2 more authors

   In International Conference on Medical Image Computing and Computer-Assisted Intervention, 2025

   Abs arXiv

   Vision-Language Foundation Models (VLFM) have shown a tremendous increase in performance in terms of generating high-resolution, photorealistic natural images. While VLFMs show a rich understanding of semantic content across modalities, they often struggle with fine-grained alignment tasks that require precise correspondence between image regions and textual descriptions a limitation in medical imaging, where accurate localization and detection of clinical features are essential for diagnosis and analysis. To address this issue, we propose a multi-stage architecture where a pre-trained VLFM provides a cursory semantic understanding, while a reinforcement learning (RL) algorithm refines the alignment through an iterative process that optimizes for understanding semantic context. The reward signal is designed to align the semantic information of the text with synthesized images. We demonstrate the effectiveness of our method on a medical imaging skin dataset where the generated images exhibit improved generation quality and alignment with prompt over the fine-tuned Stable Diffusion. We also show that the synthesized samples could be used to improve disease classifier performance for underrepresented subgroups through augmentation.

2024

1. IEEE RA-L 2024

   

   Toward reliable human pose forecasting with uncertainty

   Saeed Saadatnejad, Mehrshad Mirmohammadi, Matin Daghyani, and 6 more authors

   IEEE Robotics and Automation Letters, 2024

   Abs arXiv Code

   Recently, there has been an arms race of pose forecasting methods aimed at solving the spatio-temporal task of predicting a sequence of future 3D poses of a person given a sequence of past observed ones. However, the lack of unified benchmarks and limited uncertainty analysis have hindered progress in the field. To address this, we first develop an open-source library for human pose forecasting, including multiple models, supporting several datasets, and employing standardized evaluation metrics, with the aim of promoting research and moving toward a unified and consistent evaluation. Second, we devise two types of uncertainty in the problem to increase performance and convey better trust: 1) we propose a method for modeling aleatoric uncertainty by using uncertainty priors to inject knowledge about the pattern of uncertainty. This focuses the capacity of the model in the direction of more meaningful supervision while reducing the number of learned parameters and improving stability; 2) we introduce a novel approach for quantifying the epistemic uncertainty of any model through clustering and measuring the entropy of its assignments. Our experiments demonstrate up to 25% improvements in forecasting at short horizons, with no loss on longer horizons on Human3.6M, AMSS, and 3DPW datasets, and better performance in uncertainty estimation. The code is available online at https://github.com/vita-epfl/UnPOSed.