WSPLIN

pytorch implementation of Weakly Supervised Patch Label Inference Networks for Efficient Pavement Distress Detection and Recognition in the Wild

For more details of this task, See Pavement Distress Classification.

Usage

Crack500-PDD

Train IOPLIN

# IOPLIN use the pretrained efficientnet_b3 weight to init the model
python3 main.py --data-path=$DATA_PATH --output=$OUTPUT_PATH --project=wsplin --cfg ../configs/crack500/crack500_effi_b3.yaml ../configs/crack500/crack500_ioplin.yaml --title=ioplin_crack500 --opts MODEL.BACKBONE_INIT $PRETRAINED_WEIGHT_PATH

Train STN

python3 main.py --data-path=$DATA_PATH --output=$OUTPUT_PATH --project=wsplin --cfg ../configs/crack500/crack500_effi_b3.yaml ../configs/stn_effi_b3.yaml --title=stn_crack500
# you can change the $STN_NAME to train different stn
# $STN_NAME = {stn_1_bn,stn_1,stn_2,stn_3}
python3 main.py --data-path=$DATA_PATH --output=$OUTPUT_PATH --project=wsplin --cfg ../configs/crack500/crack500_effi_b3.yaml ../configs/stn_effi_b3.yaml --title=stn_crack500 --opts MODEL.NAME $STN_NAME

Train Other Baselines

# you can change the config file to train different baselines
python3 main.py --data-path=$DATA_PATH --output=$OUTPUT_PATH --project=wsplin --cfg ../configs/crack500/crack500_effi_b3.yaml --title=effi_b3_crack500

Train WSPLIN-IP

# WSPLIN use the pretrained efficientnet_b3 weight to init the model
python3 main.py --data-path=$DATA_PATH --output=$OUTPUT_PATH --project=wsplin --cfg ../configs/crack500/crack500_effi_b3.yaml ../configs/crack500/crack500_wsplin.yaml --title=wsplin_ip_crack500 --opts MODEL.BACKBONE_INIT $PRETRAINED_WEIGHT_PATH

CQU-BPDD

These examples are in the I-DET setting. For other settings, please change the config file.

Train IOPLIN

# IOPLIN use the pretrained efficientnet_b3 weight to init the model
python3 main.py --data-path=$DATA_PATH --output=$OUTPUT_PATH --project=wsplin --cfg ../configs/baseline/effi_b3_1det.yaml ../configs/ioplin.yaml --title=ioplin --opts MODEL.BACKBONE_INIT $PRETRAINED_WEIGHT_PATH

Train STN

python3 main.py --data-path=$DATA_PATH --output=$OUTPUT_PATH --project=wsplin --cfg ../configs/baseline/effi_b3_1det.yaml ../configs/stn_effi_b3.yaml --title=stn 
# you can change the $STN_NAME to train different stn
# $STN_NAME = {stn_1_bn,stn_1,stn_2,stn_3}
python3 main.py --data-path=$DATA_PATH --output=$OUTPUT_PATH --project=wsplin --cfg ../configs/baseline/effi_b3_1det.yaml ../configs/stn_effi_b3.yaml --title=stn --opts MODEL.NAME $STN_NAME

Train Other Baselines

# effi_b3
python3 main.py --data-path=$DATA_PATH --output=$OUTPUT_PATH --project=wsplin --cfg ../configs/baseline/effi_b3_1det.yaml --title=effi_b3 
# you can change the model name $MODEL_NAME_TIMM to train different baselines. the model name can be referd to timm repo
python3 main.py --data-path=$DATA_PATH --output=$OUTPUT_PATH --project=wsplin --cfg ../configs/baseline/effi_b3_1det.yaml --title=other_baseline --opts MODEL.NAME $MODEL_NAME_TIMM

Train WSPLIN-IP

# WSPLIN use the pretrained efficientnet_b3 weight to init the model
python3 main.py --data-path=$DATA_PATH --output=$OUTPUT_PATH --project=wsplin --cfg ../configs/baseline/effi_b3_1det.yaml ../configs/wsplin_1det.yaml --title=wsplin --opts MODEL.BACKBONE_INIT $PRETRAINED_WEIGHT_PATH

Train WSPLIN-SW

# WSPLIN use the pretrained efficientnet_b3 weight to init the model
python3 main.py --data-path=$DATA_PATH --output=$OUTPUT_PATH --project=wsplin --cfg ../configs/baseline/effi_b3_1det.yaml .../configs/wsplin_1det.yaml --title=wsplin --opts MODEL.BACKBONE_INIT $PRETRAINED_WEIGHT_PATH DATA.IS_IP False NUM_PATCHES 12

Train WSPLIN-SS

# WSPLIN use the pretrained efficientnet_b3 weight to init the model
python3 main.py --data-path=$DATA_PATH --output=$OUTPUT_PATH --project=wsplin --cfg ../configs/baseline/effi_b3_1det.yaml ../configs/wsplin_1det.yaml --title=wsplin --opts MODEL.BACKBONE_INIT $PRETRAINED_WEIGHT_PATH WSPLIN.SPARSE_RATIO 0.5