train.sh

MODEL_DIR=trained_models

MODEL_PATH=$MODEL_DIR/weakalign.pth.tar

GPU=0
NUM_WORKERS=4
BATCH_SIZE=4
LEARNING_RATE=5e-8
EPOCH=60

DATASET=pf-pascal

MATCH_LOSS=True
CYCLE_LOSS=True
TRANS_LOSS=False
COSEG_LOSS=False
TASK_LOSS=False

W_MATCH=1.0
W_CYCLE=1.0
W_TRANS=0.0
W_COSEG=0.0
W_TASK=0.0

python train.py \
    --model $MODEL_PATH \
    --training-dataset $DATASET \
    --num-epochs $EPOCH \
    --lr $LEARNING_RATE \
    --gpu $GPU \
    --num-workers $NUM_WORKERS \
    --batch-size $BATCH_SIZE \
    --result-model-dir $MODEL_DIR \
    --match-loss $MATCH_LOSS \
    --cycle-loss $CYCLE_LOSS \
    --trans-loss $TRANS_LOSS \
    --coseg-loss $COSEG_LOSS \
    --task-loss $TASK_LOSS \
    --w-match $W_MATCH \
    --w-cycle $W_CYCLE \
    --w-trans $W_TRANS \
    --w-coseg $W_COSEG \
    --w-task $W_TASK \