DDANet: Dual Decoder Attention Network for Automatic Polyp Segmentation

Authors: Nikhil Kumar Tomar, Debesh Jha, Sharib Ali, Håvard D. Johansen, Dag Johansen, Michael A. Riegler and Pål Halvorsen

Architecture

The proposed DDANet is fully convolutional network consists of a single encoder and dual decoders. The encoder consists of 4 encoder block whereas each decoder also consists of 4 decoder block. The encoder takes the RGB image as input which passes throughthe shared encoder and then it goes through both the decoders. The first decoder gives the segmentation mask and the second decoder gives the original input image in the grayscale format.

Quantative Results

Qualitative Results

Citation

Please cite our paper if you find the work useful:

@inproceedings{tomar2020ddanet,
  title={DDANet: Dual Decoder Attention Network for Automatic Polyp Segmentation},
  author={Tomar, Nikhil Kumar and Jha, Debesh and Ali, Sharib and Johansen, H{\aa}vard D and Johansen, Dag and Riegler, Michael A and Halvorsen, P{\aa}l},
  booktitle={ICPR International Workshop and Challenges},
  year={2021}
}

Contact

please contact [email protected] and [email protected] for any further questions.