A semantic segmentation workflow for working with TPS files
Overview
transfermodel_utils.PrepareData: prepares PNG images and PNG binary masks for model training (masks created using mask_from_tps.R and tps-oo.R)
train_transferlearning: trains a Unet model via transfer learning using the segmentation_models library
make_predictions: uses the trained model to predict the outlines of specimens and writes them to txt files
transfermodel_utils.WriteMultipletoTPS: writes a TPS file for the segmented outlines of however many specimens
utils.R has the exact same functionality as transfermodel_utils.py, plus the option to input JPGs instead of TIFFs
Citing
@misc{msamfairGitHub, Author = {Maya Samuels-Fair and Gene Hunt}, Title = {TPS Unet Segmentation}, Year = {2020}, Publisher = {GitHub}, Journal = {GitHub repository}, Howpublished = {\url{https://github.com/msamfair/TPS-Unet-segmentation}} }
References
Singhal, P (2019) unet_test.py. https://medium.com/@pallawi.ds/semantic-segmentation-with-u-net-train-and-test-on-your-custom-data-in-keras-39e4f972ec89.
Singhal, P (2019) unet_2.py. https://medium.com/@pallawi.ds/semantic-segmentation-with-u-net-train-and-test-on-your-custom-data-in-keras-39e4f972ec89.
Yakubovskiy, P (2019) segmentation_models. https://github.com/qubvel/segmentation_models.
Developed in Python 3.7, Tensorflow 1.15, R 3.6.1