Pytorch implementation of "Coarse to Fine: Progressive and Multi-Task Learning for Salient Object Detection"
- Clone this repository
git clone https://github.com/tiruss/PMNet.git
- You can install all the dependencies by
pip install -r requirements.txt
-
Download training datasets from the link. DUTS-TR is our train dataset.
-
Other datasets can download from the link [sal_eval_toolbox] Thank you for the awesome evaluation toolbox!
-
Download pretrained weight from the link. [Google Drive] [Baidu Drive] Baidu drive will be updated soon.
-
If you want CRF for postprocessing, download PyCRF from the link.
-
Run
python Test.py --weight [pretrained wieght] --input_dir [dataset name, default='DUTS-TE'] --crf [default='False']
- Pre-computed saliency maps can download from the link. [Google Drive] [Baidu Drive] Baidu drive will be updated soon.
-
DUTS-TR is our traning set for pair comparison
-
First, you need make contour of dataset
python Make_Contour.py --data_dir [Dataset name ex)'DUTS-TR']
- Run
python Train.py --img_dir [DUTS-TR img dir] --gt_dir [DUTS-TR label dir] --contour_dir [DUTS-TR contour dir] --epoch --batch_size --gpus --down_scale