A docker-powered weakly supervised learning framwork for neuron segmentation of fMOST and public BigNeuron datasets, based on DDeep3M network.
Thanks for the great work from DDeep3M!
-
build the Docker image for cuda-9 and cudnn-7 cd cuda-9.0-cudnn7-devel docker build -t cuda9-cudnn7 .
-
build the DDeep3M image cd ../ddeep3m docker build -t ddeep3m .
-
run the DDeep3M image and obtain an interactive bash prompt in the container nvidia-docker run -it ddeep3m:latest bash
-
pre-process the training data of fMOST with augmentation ./PreprocessTrainingData.m ./MOST/train/images ./MOST/train/labels ./preparation/
-
run the model with training data ./runtraining.sh --numiterations 1000 ./preparation/ ./MOST_trainout
-
predict the test data with trained model ./runprediction.sh ./MOST_trainout ./MOST/test/images ./MOST_predictout/
#License See LICENSE for DDeep3M