wanqiukong / rednet

This is an official release of the RedNet architecture.

RedNet

This repository contains the official implementation of the RedNet (Residual Encoder-Decoder Architecture). It turns out that the simple encoder-decoder structure is powerful when combined with residual learning. For further details of the network, please refer to our article RedNet: Residual Encoder-Decoder Network for indoor RGB-D Semantic Segmentation.

Dependencies:

PyTorch 0.4.0, TensorboardX 1.2 and other packages listed in requirements.txt.

Dataset

The RedNet model is trained and evaluated with the SUN RGB-D Benchmark suit. Please download the data on the official webpage, unzip it, and place it with a folder tree like this,

SOMEPATH # Some arbitrary path
├── SUNRGBD # The unzip folder of SUNRGBD.zip
└── SUNRGBDtoolbox # The unzip folder of SUNRGBDtoolbox.zip

The root path SOMEPATH should be passed to the program using the --data-dir SOMEPATH argument.

Usage:

For training, you can pass the following argument,

python RedNet_train.py --cuda --data-dir /path/to/SOMEPATH

If you do not have enough GPU memory, you can pass the --checkpoint option to enable the checkpoint container in PyTorch >= 0.4. For other configuration, such as batch size and learning rate, please check the ArgumentParser in RedNet_train.py.

For inference, you should run the RedNet_inference.py like this,

python RedNet_inference.py --cuda --last-ckpt /path/to/pretrained/model.pth -r /path/to/rgb.png -d /path/to/depth.png -o /path/to/output.png

The pre-trained weight is released here for result reproduction.

Citation

If you find this work to be helpful, please consider citing the paper,

@article{jiang2018rednet,
  title={RedNet: Residual Encoder-Decoder Network for indoor RGB-D Semantic Segmentation},
  author={Jiang, Jindong and Zheng, Lunan and Luo, Fei and Zhang, Zhijun},
  journal={arXiv preprint arXiv:1806.01054},
  year={2018}
}

License

This software is released under a creative commons license which allows for personal and research use only. For a commercial license please contact the authors. You can view a license summary here: http://creativecommons.org/licenses/by-nc/4.0/