no_tiling_artifacts

Tiling artifacts and trade-offs of feature normalization in the segmentation of large biological images at ICCV 2025

Tutorials

How to get rid of tiling artifacts

Is BatchNorm a good choice for my problem?

Experimental setups organoids

Name	norm layer	track_running_stats	whole vol percentile norm	tile-wise percentile norm	intensity augm	dropout	use running stats in training
setup_01	InstanceNorm	True	yes	no	no	no	no
setup_02	InstanceNorm	True	yes	yes	no	no	no
setup_03	InstanceNorm	False	yes	no	no	no	no
setup_04	InstanceNorm	False	yes	yes	no	no	no
setup_05	InstanceNorm	True	yes	no	no	no	no
setup_06	BatchRenorm	True	yes	no	no	no	no
setup_07	NoNorm	False	yes	no	no	no	no

Experimental setups plants

Name	norm layer	track_running_stats	whole vol percentile norm	tile-wise percentile norm	intensity augm	dropout	use running stats in training
setup_01	InstanceNorm	True	yes	no	no	no	no
setup_02	InstanceNorm	True	yes	yes	no	no	no
setup_03	InstanceNorm	False	yes	no	no	no	no
setup_04	InstanceNorm	False	yes	yes	no	no	no
setup_05	InstanceNorm	True	yes	no	no	no	no
setup_06	BatchRenorm	True	yes	no	no	no	no
setup_07	NoNorm	False	yes	no	no	no	no

Experimental setups UNETR

(same for all 3 datasets)

Name	norm layer	track_running_stats	whole vol percentile norm	tile-wise percentile norm	intensity augm	dropout	use running stats in training
setup_01	InstanceNorm	False	yes	no	no	no	no
setup_02	InstanceNorm	True	yes	no	no	no	no
setup_03	BatchRenorm	True	yes	no	no	no	no
setup_04	NoNorm	False	yes	no	no	no	no