Content-aware frame interpolation (CAFI): Deep Learning-based temporal super-resolution for fast bioimaging

Link to the paper: link

Usage for ZoomingSlowMo_4_Microscopy_V2.ipynb

1. Replace /home/user2/project/CAFI to your own path for the whole project including ZoomingSlowMo_4_Microscopy_V2.ipynb and ZS4Mic/load_functions.
2. This script is already adapted for Pytorch 1.11 and later. Run python ZS4Mic/codes/models/modules/dcn/setup.py build develop in the terminal to build the DCN module.
3. Run python stack_tiff.py "path/to/tiff/dir" "ZS4Mic/demo/dirname/filename.tif" to stack the tiff files into a single file.
4. Run the notebook ZoomingSlowMo_4_Microscopy_V2.ipynb to use the Zooming SlowMo model for microscopy. Only need to run 4. Training to train the model, and 5. Perform Interpolation and/or Lateral Image upscaling to perform the interpolation.
5. Run python split_tiff.py "ZS4Mic/demo/dirname/filename.tif" "path/to/tiff/dir" to split the tiff file into individual frames.

What is this?

Content-aware frame interpolation (CAFI) provides a Deep Learning-based temporal super-resolution for fast bioimaging. It increases the frame rate of any microscope modality by interpolating an image in between two consecutive images via “intelligent” interpolation, providing a 2x increase in temporal or/and axial resolution. Here we provide the modified repositories of DAIN and Zooming SlowMo used in the CAFI 4 Microscopy Google Colab notebooks.

Want to see a short video demonstration and user tutorials?

Demonstration Video	Tutorial Video CAFI (DAIN)	Tutorial Video CAFI (ZS)

Links to the notebooks and other sources

DAIN 4 Microscopy: |

Original Github of DAIN | Source Paper 1

ZoomingSlowMo 4 Microscopy |

Original Github of ZS | Source Paper 1 | Source Paper 2

Microscopy training and test data is available here:

How to cite this work

Martin Priessner, David C.A Gaboriau, Arlo Sheridan, Tchern Lenn, Jonathan R. Chubb, Uri Manor, Ramon Vilar, and Romain F. Laine

Content-aware frame interpolation (CAFI): Deep Learning-based temporal super-resolution for fast bioimaging. bioRxiv, 2021. DOI: https://doi.org/10.1101/2021.11.02.466664