Skip to content

Latest commit

 

History

History
58 lines (33 loc) · 2.56 KB

README.md

File metadata and controls

58 lines (33 loc) · 2.56 KB
Raw

RaP-Net

RaP-Net apply Region-wise weight, reflecting the invariability of regions, to re-weight Point-wise reliability of each pixel and extract local features for robust indoor localization. Technical details are described in this paper (under review)

Dongjiang Li, Jinyu Miao, Xuesong Shi, Yuxin Tian, Qiwei Long, Tianyu Cai, Ping Guo, Hongfei Yu, Wei Yang, Haosong Yue, Qi Wei, Fei Qiao, "RaP-Net: A Region-wise and Point-wise Weighting Network to Extract Robust Features for Indoor Localization," arXiv preprint arXiv:2012.00234, 2020.

If you use RaP-net in an academic work, please cite:

@article{li2020rapnet,
      title={RaP-Net: A Region-wise and Point-wise Weighting Network to Extract Robust Features for Indoor Localization}, 
      author={Dongjiang Li and Jinyu Miao and Xuesong Shi and Yuxin Tian and Tianyu Cai and Qiwei Long and Ping Guo and Hongfei Yu and Wei Yang and Haosong Yue and Qi Wei and Fei Qiao},
      journal={arXiv preprint arXiv:2012.00234},
      year={2020}
}

The code to hard detection is customized from D2-Net

1.Prerequisites

Python 3.7 is recommended for running our code. Conda can be used to install the required packages:

conda install pytorch torchvision cudatoolkit=10.0 -c pytorch
conda install h5py imageio imagesize matplotlib numpy scipy tqdm shutil

2. Models

The off-the-shelf weights of basenet and overall RaP-Net are also provided in models/rapnet.basenet.pth and models/rapnet.overall.pth respectively. They can be downloaded from release 1.0.

3. Training

We train the basenet and region-wise weight in RaP-Net separately. The related code will be open-sourced later.

4. Inference

extract_features.py can be used to extract RaP-Net features for a given list of images. The output format can be either npz or mat. In either case, the feature files encapsulate three arrays:

- keypoints [N x 3] array containing the positions of keypoints x, y and the scales s. The positions follow the COLMAP format, with the X axis pointing to the right and the Y axis to the bottom.

- scores [N] array containing the activations of keypoints (higher is better).

- descriptors [N x 512] array containing the L2 normalized descriptors.

python extract_features.py --image_list /path/to/image/list/ --file_type jpg(png or jpeg) --model_file models/rapnet.overall.pth