Generating pseudo labels for MFNet.
This can also be served as a pipeline for all weakly supervised salient object detection (WSOD) methods. This code can generate class activation maps (CAMs) as well as two kinds of pseudo labels for WSOD. We sincerely hope that this will contribute to the community.
- Windows 10
- Torch 1.8.1
- CUDA 10.0
- Python 3.7.4
- other environment requirment can be found in requirments.txt
you can download ImageNet dataset from this official website.
you can download DUTS-Train dataset from this official website. Only RGB images are used in our MFNet.
you should set your training and inference dataset root in --cls_dataset_dir and --sal_dataset_dir in run_sample.py, respectively.
setting --train_cam_pass to True, and run run_sample.py.
setting --make_cam_pass to True, and run run_sample.py. Here you can get ① CAMs and ② the pixel-wise pseudo labels in root ./result/.
setting your inference dataset root in img_root in run_slic.py, and run. Here you can get ③superpixel-wise pseudo labels in root ./result/.
link: https://pan.baidu.com/s/1G-YHYsfho-rWwMv6VMFT4g. code: oipw
link: https://pan.baidu.com/s/1nowIVfeauJs6w_k4waBqOA. code: oipw
Thanks to pioneering helpful works:
- IRNet: Weakly Supervised Learning of Instance Segmentation with Inter-pixel Relations, CVPR2019, by Jiwoon Ahn et al.
- MSW: Multi-source weak supervision for saliency detection, CVPR2019, by Yu Zeng et al.
- SSSS: Single-stage Semantic Segmentation from Image Labels, CVPR2020, by Nikita Araslanov et al.