Rethinking Low-level Features for Interest Point Detection and Description

Dependency

pytorch
torchvision
cv2
tqdm

We use cuda 11.4/python 3.8.13/torch 1.10.0/torchvision 0.11.0/opencv 3.4.8 for training and testing.

Pre-trained models

We provide two versions of LANet with different structure in network_v0 and network_v1, the corresponding pre-trained models are in checkpoints.

v0: The original version used in our paper.
v1: An improved version that has a better over all performance.

Training

Download the COCO dataset:

cd datasets/COCO/
wget http://images.cocodataset.org/zips/train2017.zip
unzip train2017.zip

Prepare the training file:

python datasets/prepare_coco.py --raw_dir datasets/COCO/train2017/ --saved_dir datasets/COCO/

To train the model (v0) on COCO dataset, run:

python main.py --train_root datasets/COCO/train2017/ --train_txt datasets/COCO/train2017.txt

Evaluation

Evaluation on HPatches dataset

Download the HPatches dataset:

cd datasets/HPatches/
wget http://icvl.ee.ic.ac.uk/vbalnt/hpatches/hpatches-sequences-release.tar.gz
tar -xvf hpatches-sequences-release.tar.gz

To evaluate the pre-trained model, run:

python test.py --test_dir ./datasets/HPatches/hpatches-sequences-release

License

The code is released under the MIT license.

Citation

Please use the following citation when referencing our work:

@InProceedings{Wang_2022_ACCV,
    author    = {Changhao Wang and Guanwen Zhang and Zhengyun Cheng and Wei Zhou},
    title     = {Rethinking Low-level Features for Interest Point Detection and Description},
    booktitle = {Computer Vision - {ACCV} 2022 - 16th Asian Conference on Computer
                 Vision, Macao, China, December 4-8, 2022, Proceedings, Part {II}},
    series    = {Lecture Notes in Computer Science},
    volume    = {13842},
    pages     = {108--123},
    year      = {2022}
}

Related Projects

https://github.com/TRI-ML/KP2D