This is the code for our recent work that develops a new method for Global Pixel Transformers for Virtual Staining. The code is created and modified based upon the work from Google. The code to their work is at https://github.com/google/in-silico-labeling.

Citing bibtex for our work will be avialable upon the publishing of our paper.

We have tested this code using:

• Ubuntu 18.04
• Python 3
• NumPy
• TensorFlow
• OpenCV

Data is available at https://github.com/google/in-silico-labeling/blob/master/data.md.

python gunet\launch.py -- \
  --alsologtostderr \
  --base_directory $BASE_DIRECTORY \
  --mode EVAL_EVAL \
  --metric INFER_FULL \
  --stitch_crop_size 1500 \
  --restore_directory $(pwd)/checkpoints \
  --read_pngs \
  --dataset_eval_directory $(pwd)/data_sample/condition_b_sample \
  --infer_channel_whitelist DAPI_CONFOCAL,MAP2_CONFOCAL,NFH_CONFOCAL

In the above:

1. BASE_DIRECTORY is the working directory for the model. It will be created if it doesn't already exist, and it's where the model predictions will be written. You can set it to whatever you want.
2. alsologtostderr will cause progress information to be printed to the terminal.
3. stitch_crop_size is the size of the crop for which we'll perform inference. If set to 1500 it may take an hour on a single machine, so try smaller numbers first.
4. infer_channel_whitelist is the list of fluorescence channels we wish to infer. For the Condition B data, this should be a subset of DAPI_CONFOCAL, MAP2_CONFOCAL, and NFH_CONFOCAL.

The code is GNU General Public licensed, as found in the LICENSE file.