How to Apply CNNs to Large Image Data — or —What can go wrong with tile-and-stitch?
- Abstract number
- 94
- Presentation Form
- Invited
- Corresponding Email
- [email protected]
- Session
- Image Data Analysis, Management and Visualisation
- Authors
- Dagmar Kainmueller (1)
- Affiliations
-
1. Janelia Research Campus
- Abstract text
Microscopy data is often large. Hence when processed with convolutional neual networks (CNNs) like the popular UNet, to cope with GPU memory constraints, the data has to be tiled into smaller pieces which are processed individually, and results are stitched back together afterwards. In this context, issues with inconsistencies at stitching seams have been reported. However, a formal analysis of the causes has been lacking. Our work shows that the potential for inconsistencies to arise is intricately tied to the shift equivariance properties of the employed CNNs. Our theoretical analysis entails simple rules for designing CNNs that are necessary to avoid inconsistencies in tile-and-stitch processing of large data.