This is the PyTorch implementation repository of our [IJCV 2021] paper: Cross-Modal Pyramid Translation for RGB-D Scene Recognition

We propose a general and flexible multi-modal learning framework, to unify the tasks of modality specific recognition and cross-modal image translation with a shared representation, thus improving its generalization power. A preliminary version of this work was presented in the conference paper: TRecgNet (https://github.com/ownstyledu/Translate-to-Recognize-Networks).

• Python 3.6 or higher
• PyTorch 1.4 or higher
• Torchvision
• TensorboardX

We provide links(https://pan.baidu.com/s/1LZIF1hlT3k0oX76Ttp660w, extraction code: g5vp) to download SUN RGB-D data in ImageFolder format and depth data has been encoded using HHA format. The code runs with the concatenated version of this data.

One can easily config all the hyper-parameters using the files in the config package. An example to run the Depth TRecgNet (depth to RGB) with multiscale image translation is shown as follows (0 refers to the gpu idx):

bash train.sh 0 sunrgbd arch=resnet18 content_arch=resnet18 multi_scale=True direction=BtoA model=trecg no_trans=False loss_types=CLS,PERCEPTUAL,AUX_CLS

Please cite the following paper if you feel this repository useful.

@article{du2021cross,
  title={Cross-Modal Pyramid Translation for RGB-D Scene Recognition},
  author={Du, Dapeng and Wang, Limin and Li, Zhaoyang and Wu, Gangshan},
  journal={International Journal of Computer Vision},
  volume={129},
  number={8},
  pages={2309--2327},
  year={2021},
  publisher={Springer}
}

@inproceedings{du2019translate,
  title={Translate-to-Recognize Networks for RGB-D Scene Recognition},
  author={Du, Dapeng and Wang, Limin and Wang, Huiling and Zhao, Kai and Wu, Gangshan},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
  pages={11836--11845},
  year={2019}
}