Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
We introduce a novel diversity-aware diagnostic for mask-conditioned diffusion models in biomedical imaging, addressing the inadequacy of standard metrics like FID/KID which fail to capture domain-specific failures. Our diagnostic, operating in a domain-validated feature space, quantifies both realism and texture diversity, identifying a critical failure mode where models collapse to low-diversity outputs late in training. This approach enables selection of checkpoints that yield synthetic data indistinguishable from real microscopy by experts and support downstream segmentation tasks with superior performance and reduced variance compared to existing methods.
Mask-conditioned diffusion models can achieve high visual realism while secretly collapsing texture diversity, a failure mode missed by standard metrics but detectable with a new diagnostic.
Mask-conditioned diffusion models have become standard for generating labeled training data in biomedical imaging. But the tools used to evaluate them were built for a different problem. Standard metrics like FID and KID score images through ImageNet-trained features. Those features do not transfer to microscopy, leaving the metrics poorly calibrated for specialized domains. Even with a domain-appropriate feature space, a single blended number can still hide a real failure: a model can produce visually plausible images that still transfer poorly to downstream tasks. We identify a specific failure mode: Late in training, realism keeps improving while texture diversity collapses. The model settles onto a fixed color palette, converging to a high-similarity, low-diversity state that standard metrics do not penalize. Therefore, we propose a diversity-aware diagnostic in a domain-validated feature space. It combines a realism axis based on inter-similarity to real images with a diversity axis based on intra-similarity among generated samples. This diagnostic selects a checkpoint that domain experts cannot reliably distinguish from real recordings in a forced-choice study. It also yields useful downstream segmentation: a segmenter trained only on synthetically labeled data reaches a median skeletonised IoU comparable to one trained on real data, at lower cross-fold variance, and clearly ahead of the best available public alternative in this domain, a parametric renderer. We also reproduce a data-efficient hyperparameter-transfer experiment from prior work, tuning several foundation segmenters on a small labeled subset and evaluating on real images: transfer is stronger with our data. We release DiffuMT on HuggingFace, including the triplet dataset, code to reproduce the downstream-utility validation, and a standalone diagnostic tool for evaluating mask-conditioned diffusion models.