pre-train.sh

#!/bin/sh

export MASTER_ADDR=$(/usr/sbin/ip a show dev bond0 | grep inet | cut -d " " -f 6 | cut -d "/" -f 1)
export NGPUS=8
export NPERNODE=4


# If the model is pre-trained with 3D-OFDB....

out_dir=outputs/3d-ofdb

mpiexec -npernode ${NPERNODE} -np ${NGPUS} python -B main.py data=grayimagefolder \
                data.baseinfo.name=3d-ofdb-1k \
                data.baseinfo.train_imgs=1000 data.baseinfo.num_classes=1000 \
                data.trainset.root=../../../../../../dataset/3D-OFDB-1000/image \
                data.loader.batch_size=32 data.transform.no_aug=False data.transform.auto_augment=rand-m9-mstd0.5-inc1 \
                data.transform.re_prob=0.5 data.transform.color_jitter=0.4 data.transform.hflip=0.5 data.transform.vflip=0.5 \
                data.transform.scale=[0.08,1.0] data.transform.ratio=[0.75,1.3333] data.mixup.prob=1.0 data.mixup.mixup_alpha=0.8 \
                data.mixup.cutmix_alpha=1.0 data.mixup.switch_prob=0.5 model=vit \
                model.arch.model_name=vit_tiny_patch16_224 model.optim.learning_rate=0.001 \
                model.scheduler.args.warmup_epochs=5 logger.group=cpmpare_instance_augmentation \
                logger.save_epoch_freq=10000 epochs=80000 mode=pretrain \
                output_dir=${out_dir}



# If the model is pre-trained with 2D-OFDB....
# out_dir=outputs/2d-ofdb
# mpiexec -npernode ${NPERNODE} -np ${NGPUS} python -B main.py data=grayimagefolder \
#                 data.baseinfo.name=2d-ofdb-1k \
#                 data.baseinfo.train_imgs=1000 data.baseinfo.num_classes=1000 \
#                 data.trainset.root=../../../../../../dataset/2D-OFDB-1000/image \
#                 data.loader.batch_size=32 data.transform.no_aug=False data.transform.auto_augment=rand-m9-mstd0.5-inc1 \
#                 data.transform.re_prob=0.5 data.transform.color_jitter=0.4 data.transform.hflip=0.5 data.transform.vflip=0.5 \
#                 data.transform.scale=[0.08,1.0] data.transform.ratio=[0.75,1.3333] data.mixup.prob=1.0 data.mixup.mixup_alpha=0.8 \
#                 data.mixup.cutmix_alpha=1.0 data.mixup.switch_prob=0.5 model=vit \
#                 model.arch.model_name=vit_tiny_patch16_224 model.optim.learning_rate=0.001 \
#                 model.scheduler.args.warmup_epochs=5 logger.group=cpmpare_instance_augmentation \
#                 logger.save_epoch_freq=10000 epochs=80000 mode=pretrain \
#                 output_dir=${out_dir}