BigCell: Generating Gigapixel Whole-Slide Images
BigCell generates coherent gigapixel histopathology slides, and synthetic samples raise few-shot classification by up to 18%.
BigCell generates coherent whole-slide histopathology images at native gigapixel resolution, unlike prior patch generators limited to at most 1024x1024 tiles. A flow-matching transformer is trained at 2048x2048 on more than 120,000 public WSIs across organs, then a sliding-window pipeline conditioned on pathology foundation-model embeddings synthesizes up to 32,768x32,768 pixels. Adaptive scheduling speeds generation by more than 3x, a separate model enables text-to-image synthesis at 8k resolution, and synthetic augmentation improves few-shot slide classification by up to 18%.
- Flow-matching transformer trained at 2048x2048 on over 120,000 WSIs.
- Sliding-window inference produces images up to 32,768x32,768 pixels.
- Adaptive scheduling speeds generation more than 3x.
- Synthetic augmentation lifts few-shot slide classification by up to 18%.
From the original article
1 min read at huggingface.co
Whole-slide images (WSIs) are gigapixel-scale digital scans of histopathology tissue and serve as the primary basis for cancer diagnosis. Existing histopathology generative models operate at the patch scale, producing fixed-size tiles of at most 1024times1024 pixels which encompass only a small fraction of the WSI.
By Srikar Yellapragada, Alexandros Graikos, Zilinghan Li · 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.