coco_train.sh

EXPNAME=$1
SAVEDIR=checkpoints/coco/${EXPNAME}
IMAGENET_PRETRAIN=weight/R-101.pkl                           
IMAGENET_PRETRAIN_TORCH=weight/resnet101-5d3b4d8f.pth  


python3 train_net.py --num-gpus 4 --config-file configs/coco/base.yaml     \
    --opts MODEL.WEIGHTS ${IMAGENET_PRETRAIN}                                         \
           OUTPUT_DIR ${SAVEDIR}/base


python3 tools/model_surgery.py --dataset coco --method randinit                        \
    --src-path ${SAVEDIR}/base/model_final.pth                         \
    --save-dir ${SAVEDIR}/base
BASE_WEIGHT=${SAVEDIR}/base/model_reset_surgery.pth


for seed in 0 
do
    for shot in 1 2 3 5 10 30
    do
        python3 tools/create_config.py --dataset coco14 --config_root configs/coco     \
            --shot ${shot} --seed ${seed} --setting 'gfsod'
        CONFIG_PATH=configs/coco/mfdc_gfsod_novel_${shot}shot_seed${seed}.yaml
        OUTPUT_DIR=${SAVEDIR}/mfdc_gfsod_novel/tfa-like/${shot}shot_seed${seed}
        python3 train_net.py --num-gpus 4 --config-file ${CONFIG_PATH}                      \
            --opts MODEL.WEIGHTS ${BASE_WEIGHT} OUTPUT_DIR ${OUTPUT_DIR}               \
                   TEST.PCB_MODELPATH ${IMAGENET_PRETRAIN_TORCH}
        rm ${CONFIG_PATH}
        rm ${OUTPUT_DIR}/model_final.pth
    done
done