Figure 5
From: Flexible Learning-Free Segmentation and Reconstruction of Neural Volumes
An example image from sSEM Subvolume 2 and the corresponding APEX2 labeled process segmentation results of various methods. The first row shows the result processed by the FLoRIN pipeline, and results of thresholding with NDNT 3D and 2D without any pre/post-processing. The 3D U-Net result was of poor quality, and the original U-Net result contains many merge errors. Similarly, other methods did not produce acceptable results. The Yen48 result comes close to being useful, but this method did not show the same performance in the other imaging modalities. Methods labeled with asterisks were applied over 3D blocks of the volume, which in extreme cases resulted in the white “tiles” in some of the images.
