Training code for SenNet + HOA - Hacking the Human Vasculature in 3D (1st place solution).
It's recommended to save memmap files to shared memory(/dev/shm) to accelerate 3d rotation.
# generate memmap files
python prepare_data.py \
-s [dataset path] \
-o [save dir]
# train
CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python train.py \
--memmap_dir [memmap path]