Jingfeng-Tang/CB-SCTC

Accepted by IEEE TCSVT

[TCSVT] Cross-block Sparse Class Token Contrast for Weakly Supervised Semantic Segmentation

Code

Fig.1 - The architecture of the proposed WSSS framework(CB-SCTC)

Prerequisite

• Ubuntu 20.04, with Python 3.7 and the following python dependencies.

pip install -r requirements.txt

• Download the PASCAL VOC 2012 training and validation sets.

• Download the PASCAL VOC 2012 test set.

• Download the PASCAL VOC 2012 SBD set.

• Download the MS COCO 2014 training set.

• Download the MS COCO 2014 validation set.

• Download the MS COCO 2014 training and validation annotations.

• Download the segNet pre-trained weight.

Run for PASCAL VOC 2012 dataset

bash run_for_PASCALVOC2012.sh

Run for MS COCO 2014 dataset

bash run_for_MSCOCO2014.sh

ToDO

• update latest code
• add MCTformer+(w/ours) code

Contact

If you have any questions, you can either create issues or contact me by email tangjingfeng@stmail.ujs.edu.cn

Acknowledgement

We heavily borrowed MCTFormerV1 to construct our backbone. Many thanks for their brilliant work!