MouseLand
[FEATURE] patch-wise diameter estimation for large or heterogeneous images
Is your feature request related to a problem? Please describe. I'm working with large, heterogeneous tissue images (e.g., tumor samples) where cell sizes vary significantly across regions. The current global diameter estimation may not reflect local variation and can lead to inaccurate segmentation. Additionally, processing large images often exceeds memory limits.
Describe the solution you'd like Enable patch-wise diameter estimation: divide large images into smaller tiles, estimate diameter per tile, and aggregate (e.g., median or mean) to obtain a more robust global diameter. This would improve segmentation accuracy and reduce memory usage.
Describe alternatives you've considered None
Additional context At distributed API, diameter estimation appears to be global without support for region-wise estimation. Patch-wise diameter estimation could be a better solution?🤔🤔🤔
Hi @ShuaibingRen does CP4 work with your samples? It was trained to work with more heterogenous sizes of objects. When you run CP4 without a diameter set, it just uses the full resolution image.
