SphereFlow imputes missing modalities via hyperspherical transport
Original title: SphereFlow: Missing Modality Imputation via Geometric Transport on Hypersphere
SphereFlow projects modalities to shared hypersphere for short-range cross-modal flow, achieving state-of-the-art on BraTS MRI imputation and closing 80% of segmentation gap.
Existing methods force long jumps between modality manifolds. SphereFlow uses frozen self-supervised encoder to project to hypersphere. Flow matching with geometry-aware modifications resolves noise and interpolation conflicts. Experiments show SOTA generation fidelity and closes 80% gap to full-modality oracle in downstream brain tumor segmentation.
- Projects modalities to shared hyperspherical latent space
- Resolves Gaussian noise mismatch and linear interpolation conflicts
- State-of-the-art on BraTS multi-modal MRI imputation
- Closes 80% of gap to full-modality oracle in segmentation
- Organizations
- huggingface
- AI models
- SphereFlow
From the original article
1 min read at huggingface.co
Missing modality imputation aims to synthesize unobserved modalities from observed ones. Existing methods treat each modality as living on its own latent manifold, forcing the generator to learn a long, unstructured jump between them.
By Huihan Wang, Juncheng Wang, Yuxiang Feng · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.