Demo of Depth-Adapted CNN for RGB-D Cameras (ACCV2020 Oral)

aCNN.py contains the basic functions for computing depth adapted sampling position.

model.py shows a VGG-16 adapted to depth.

We also propose an extention of our conferece paper which is available here:

Depth-Adapted CNNs for RGB-D Semantic Segmentation

We extensively compare our approach with the concurrent SConv TIP21 and ShapeConv ICCV21.

We show that our approach can achieve better performance when all methods are under the same baseline.

The source code will be released soon.

We show that depth offset can be useful for other segmentation task such as salient object dectection.

If you find this repo useful, please consider citing:

@inproceedings{wu2020depth,
  title={Depth-adapted CNN for RGB-D cameras},
  author={Wu, Zongwei and Allibert, Guillaume and Stolz, Christophe and Demonceaux, C{\'e}dric},
  booktitle={Proceedings of the Asian Conference on Computer Vision},
  year={2020}
}

@article{wu2022depth,
  title={Depth-Adapted CNNs for RGB-D Semantic Segmentation},
  author={Wu, Zongwei and Allibert, Guillaume and Stolz, Christophe and Ma, Chao and Demonceaux, C{\'e}dric},
  journal={arXiv preprint arXiv:2206.03939},
  year={2022}
}