voc_train.sh

#!/usr/bin/env bash

EXP_NAME=$1
SAVE_DIR=checkpoints/voc/${EXP_NAME}
IMAGENET_PRETRAIN=weight/R-101.pkl                            
IMAGENET_PRETRAIN_TORCH=weight/resnet101-5d3b4d8f.pth  
SPLIT_ID=$2


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


python3 tools/model_surgery.py --dataset voc --method randinit                                \
    --src-path ${SAVE_DIR}/base${SPLIT_ID}/model_final.pth                    \
    --save-dir ${SAVE_DIR}/base${SPLIT_ID} 

BASE_WEIGHT=${SAVE_DIR}/base${SPLIT_ID}/model_reset_surgery.pth


for seed in 0 
do
    for shot in 1 2 3 5 10  
    do
        python3 tools/create_config.py --dataset voc --config_root configs/voc               \
            --shot ${shot} --seed ${seed} --setting 'gfsod' --split ${SPLIT_ID}
        CONFIG_PATH=configs/voc/mfdc_gfsod_novel${SPLIT_ID}_${shot}shot_seed${seed}.yaml
        OUTPUT_DIR=${SAVE_DIR}/mfdc_gfsod_novel${SPLIT_ID}/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