colorblank/SiamDLT

Similarity Learning for Dense Label Transfer pytorch implementation

SiamDLT (Similarity Learning for Dense Label Transfer)

a pytorch implementation of the paper Similarity Learning for Dense Label Transfer

Example

Usage

---

requirement:

• pytorch 0.4.0
• python 3.6
• GPU support
• DAVIS2017 dataset

Training(from deeplab pretrained model)

1. download deeplab pretrained model which is support by pytorch-deeplab-resnet , put it in pretrained/deeplab.pth
2. modify DAVIS_PATH
3. run python main.py

Training(from checkpoint)

0. pretrained model will be available soon....
1. modify DAVIS_PATH
2. modify ckpt_file
3. run python main.py

Inference

1. modify ckpt_file
2. run python eval.py
3. results will saved in result/ , you can modify here to see more results

Acknowledgment

The implementation of deeplab is heavily borrowed from pytorch-deeplab-resnet

Reference

• Similarity Learning for Dense Label Transfer

Citing

If you find this code useful, please cite:

@article{DAVIS2018-Interactive-2nd,
          author = {M. Najafi, V. Kulharia, T. Ajanthan, P. H. S. Torr},
          title = {Similarity Learning for Dense Label Transfer},
          journal = {The 2018 DAVIS Challenge on Video Object Segmentation - CVPR Workshops},
          year = {2018}
        }