For more meaningful and rigorous research for Weakly-Supervised Semantic Segmentation (WSSS).
DATASET=voc
SAL_DIR=saliency_maps/VOC2012_MOVE
ATM_DIR=activation_maps/CAM
SAVE_DIR=pseudo_labels/VOC2012_CAM_MOVE
ATM_THRESH=0.1
IGNORE_LABEL=True
WO_SAL=False
WORKERS=8
python3 run.py --sal_dir ${SAL_DIR} --atm_dir ${SAL_DIR --save_dir ${SAVE_DIR} --dataset ${DATASET} --atm_thresh ${ATM_THRESH} --ignore_label ${IGNORE_LABEL} --wo_sal ${WO_SAL} --workers ${WORKERS}
or
we provide the recommended setting for each WSSS method in scripts/
.
bash scripts/cam_voc12.sh
bash scripts/eps_voc12.sh
bash scripts/cam_coco14.sh
bash scripts/l2g_coco14.sh
We provide the download link for the activation map from below WSSS method.
We provide the download link for the below saliency map.
-
VOC2012
- DSS_MSRA
- DSS_MSRA_HKU
- DSS_DUTS
- DSS_COCO
- DSS_COCO20
- PFAN_MSRA
- PFAN_MSRA_HKU
- PFAN_DUTS
- PFAN_COCO
- PFAN_COCO20
- PoolNet_MSRA
- PoolNet_MSRA_HKU
- PoolNet_DUTS
- PoolNet_COCO
- PoolNet_COCO20
- VST_MSRA
- VST_MSRA_HKU
- VST_DUTS
- VST_COCO
- VST_COCO20
- DeepUSPS
- MOVE
-
COCO2014
- DSS_MSRA
- DSS_MSRA_HKU
- DSS_DUTS
- PFAN_MSRA
- PFAN_MSRA_HKU
- PFAN_DUTS
- PoolNet_MSRA
- PoolNet_MSRA_HKU
- PoolNet_DUTS
- VST_MSRA
- VST_MSRA_HKU
- VST_DUTS
- DeepUSPS
- MOVE
Please make an issue first with your downloadable link of activation maps or saliency maps which is compatible with our framework.