pretrain.sh

set -x
# Set the path to save checkpoints
OUTPUT_DIR='exp/pretrain/mae_800ep_bs4096_base_size224_patch16_mask75_decdepth8_decdim512_norm_pos2d_mocoinit'
# path to imagenet-1k train set
DATA_PATH='./data/lmdb/train.lmdb'


# batch_size can be adjusted according to the graphics card
OMP_NUM_THREADS=1 python -m torch.distributed.launch --nproc_per_node=8 run_mae_pretraining.py \
        --data_path ${DATA_PATH} \
        --normlize_target True \
        --mask_ratio 0.75 \
        --model pretrain_mae_base_patch16_224 \
        --batch_size 512 \
	  --num_workers 8 \
        --opt adamw \
        --opt_betas 0.9 0.95 \
        --warmup_epochs 40 \
        --epochs 800 \
        --save_ckpt_freq 80 \
        --output_dir ${OUTPUT_DIR}


# # Set the path to save checkpoints
# OUTPUT_DIR='exp/e2e_finetune/mae_800ep_bs4096_base_size224_patch16_mask75_decdepth8_decdim512_unnorm_pos2d_mocoinit_v2'
# # path to imagenet-1k set
# DATA_PATH='./datasets/lmdb'
# # path to pretrain model
# MODEL_PATH='./exp/pretrain/mae_800ep_bs4096_base_size224_patch16_mask75_decdepth8_decdim512_unnorm_pos2d_mocoinit/checkpoint-799.pth'

# # batch_size can be adjusted according to the graphics card
# OMP_NUM_THREADS=1 python -u -W ignore \
#     -m torch.distributed.launch --nproc_per_node=8 run_class_finetuning.py \
#     --model vit_base_patch16_224 \
#     --data_path ${DATA_PATH} \
#     --finetune ${MODEL_PATH} \
#     --output_dir ${OUTPUT_DIR} \
#     --batch_size 128 \
#     --opt adamw \
#     --opt_betas 0.9 0.999 \
#     --weight_decay 0.05 \
#     --epochs 100 \
#     --reprob 0 \
#     --use_cls \
#     --dist_eval