README.md

Z-upscaling

This repository contains the source code for our paper:

Z-upscaling: Optical Flow Guided Frame Interpolation for Isotropic Reconstruction of 3D EM Volumes
IEEE International Symposium on Biomedical Imaging (ISBI), 2025
Fisseha A. Ferede, Ali Khalighifar, Jaison John, Krishnan Venkataraman, Khaled Khairy

Demo

Input Downsampled Volume

Upscaled Volume (Z × 8)

Ground Truth Volume

Evaluation

--pattern represents a path where 3D image volumes in .tiff format and/or sub-directories containing z-slices of a given volume in a sorted order are located. --model_path a path where the model used for evaluation is located (Download pretrained models Saved models). --outputfile a path where the upsampled volumes will be saved. --times_to_interpolate isotropizing factor by which the volume input is upscaled. If it's in 2^n order, it invokes recursive spatial interpolation to achieve the target number of frames, if not it will invoke the interpolation to the nearest 2^n order and apply bicubic interpolation to upsample or downsample. --output_volume if true, isotropized volume will be saved in .tiff format. --remove_sliced_volumes if true, removes the intermediate 2D slices generated from a 3D .tiff volume input.

python3 /Z-upscaling-main/eval/interpolator_cli.py \
   --pattern "/Z-upscaling-main/Demo/*" \
   --model_path /Z-upscaling-main/ModelPaths/test_run_ft_em_/saved_model_2M \
   --outputfile /Z-upscaling-main/Demo_out \
   --times_to_interpolate 8 \
   --output_volume "True" \
   --remove_sliced_volumes "False"

Cite

If you find this work useful please cite:

@article{ferede2024z,
  title={Z-upscaling: Optical Flow Guided Frame Interpolation for Isotropic Reconstruction of 3D EM Volumes},
  author={Ferede, Fisseha A and Khalighifar, Ali and John, Jaison and Venkataraman, Krishnan and Khairy, Khaled},
  journal={arXiv preprint arXiv:2410.07043},
  year={2024}
}