Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2503.19545 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866915213116178432 |
|---|---|
| author | Buglakova, Elena Archit, Anwai D'Imprima, Edoardo Mahamid, Julia Pape, Constantin Kreshuk, Anna |
| author_facet | Buglakova, Elena Archit, Anwai D'Imprima, Edoardo Mahamid, Julia Pape, Constantin Kreshuk, Anna |
| contents | Segmentation of very large images is a common problem in microscopy, medical imaging or remote sensing. The problem is usually addressed by sliding window inference, which can theoretically lead to seamlessly stitched predictions. However, in practice many of the popular pipelines still suffer from tiling artifacts. We investigate the root cause of these issues and show that they stem from the normalization layers within the neural networks. We propose indicators to detect normalization issues and further explore the trade-offs between artifact-free and high-quality predictions, using three diverse microscopy datasets as examples. Finally, we propose to use BatchRenorm as the most suitable normalization strategy, which effectively removes tiling artifacts and enhances transfer performance, thereby improving the reusability of trained networks for new datasets. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2503_19545 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Tiling artifacts and trade-offs of feature normalization in the segmentation of large biological images Buglakova, Elena Archit, Anwai D'Imprima, Edoardo Mahamid, Julia Pape, Constantin Kreshuk, Anna Computer Vision and Pattern Recognition Segmentation of very large images is a common problem in microscopy, medical imaging or remote sensing. The problem is usually addressed by sliding window inference, which can theoretically lead to seamlessly stitched predictions. However, in practice many of the popular pipelines still suffer from tiling artifacts. We investigate the root cause of these issues and show that they stem from the normalization layers within the neural networks. We propose indicators to detect normalization issues and further explore the trade-offs between artifact-free and high-quality predictions, using three diverse microscopy datasets as examples. Finally, we propose to use BatchRenorm as the most suitable normalization strategy, which effectively removes tiling artifacts and enhances transfer performance, thereby improving the reusability of trained networks for new datasets. |
| title | Tiling artifacts and trade-offs of feature normalization in the segmentation of large biological images |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2503.19545 |