RefFree: Reference-Free 3D Reconstruction of brain dissection photographs with ML

Official Implementation for "Reference-Free 3D Reconstruction of Brain Dissection Photographs with Machine Learning"

RefFree is a deep learning-based ex vivo brain dissection photograph reconstruction pipeline. It consists of two stages: (1) a MNI coordinate prediction neural network trained with a dissection photograph synthetic data engine, and (2) an MNI coordinate-based reconstruction algorithm as shown below.

TODO:

• Release data processing code

👉 Installation

Run the following

conda create -n reffree python==3.9
conda activate reffree

cd reffree
pip install .

👉 Data preprocessing

👉 Training

Before training, create a train_config.py file from configs/train_config_example.py. Update the directory setting in train_config.py.

Train the model with the following command line

python src/reffree/train.py -e [EXP_ID] --exp_folder [EXP_FOLDER] --batch_size 2 --slice_num 32 --epoch 100 --train_config ./configs/train_config.py