Mischa Dombrowski
11 Papers
Mischa Dombrowski is an academic researcher. The author has contributed to research in topics: Computer science & Engineering. The author has an hindex of 1, co-authored 7 publications.
Chat about Author
Papers
Feature-Conditioned Cascaded Video Diffusion Models for Precise Echocardiogram Synthesis
Hadrien Reynaud,Mengyun Qiao,Mischa Dombrowski,Thomas G. Day,Reza Razavi,Alberto Gomez,Paul Leeson,B. Kainz +7 more
TL;DR: In this article , the authors extend elucidated diffusion models for video modelling to generate plausible video sequences from single images and arbitrary conditioning with clinical parameters, which achieves an $R^2$ score of 93%, which is 38 points higher than recently proposed sequence to sequence generation methods.
Pay Attention: Accuracy Versus Interpretability Trade-off in Fine-tuned Diffusion Models
TL;DR: The authors showed that fine-tuning text-to-image models with learnable text encoder leads to a lack of interpretability of diffusion models, and demonstrated the interpretability by keeping the language encoder frozen, enables diffusion models to achieve state-of-the-art phrase grounding performance on certain diseases for a challenging multi-label segmentation task, without any additional training.
Quantifying Sample Anonymity in Score-Based Generative Models with Adversarial Fingerprinting
Mischa Dombrowski,B. Kainz +1 more
TL;DR: In this paper , the authors introduce a method for estimating the upper bound of the probability of reproducing identifiable training images during the sampling process by designing an adversarial approach that searches for anatomic fingerprints, such as medical devices or dermal art, which could potentially be employed to re-identify training images.
1
GRASP: Guided Residual Adapters with Sample-wise Partitioning
Felix Nützel,Mischa Dombrowski,Bernhard Kainz +2 more
TL;DR: GRASP, a novel method, addresses mode collapse in long-tail text-to-image diffusion models by partitioning samples into clusters and injecting cluster-specific residual adapters, improving FID and diversity metrics, especially for rare classes, on medical imaging datasets.
Generate to Ground: Multimodal Text Conditioning Boosts Phrase Grounding in Medical Vision-Language Models
Felix Nützel,Mischa Dombrowski,Bernhard Kainz +2 more
TL;DR: This study demonstrates that generative text-to-image diffusion models outperform discriminative methods in phrase grounding for medical imaging, achieving double the mIoU scores with a novel post-processing technique, Bimodal Bias Merging, and a domain-specific language model.