├── README.md
└── assets
    ├── network.pdf
    └── network.png


/README.md:
--------------------------------------------------------------------------------
 1 | # AdaMatcher: Adaptive Assignment for Geometry Aware Local Feature Matching
 2 | ### [Paper](https://arxiv.org/abs/2207.08427)
 3 | 
 4 | <br/>
 5 | 
 6 | > Adaptive Assignment for Geometry Aware Local Feature Matching
 7 | > Dihe Huang<sup>\*</sup>, Ying Chen<sup>\*</sup>, Yong Liu, Jianlin Liu, Shang Xu, Wenlong Wu, Yikang Ding, Fan Tang, Chengjie Wang
 8 | > CVPR 2023
 9 | 
10 | :triangular_flag_on_post: **Updates**
11 | Code has been migrated to official repository: https://github.com/TencentYoutuResearch/AdaMatcher
12 | 
13 | ![network](assets/network.png)
14 | 
15 | 
16 | 
17 | ## Installation
18 | For environment and data setup, please refer to [LoFTR](https://github.com/zju3dv/LoFTR).
19 | 
20 | 
21 | ## Run AdaMatcher
22 | 
23 | ### Download Pretrained model
24 | We have provide pretrained model in megadepth dataset, you can download it from [weights](https://drive.google.com/drive/folders/1067_GfX7i_ZLj6Sp68S3d9cofdaPlDZW?usp=share_link).
25 | 
26 | ### Download Datasets
27 | You need to setup the testing subsets of ScanNet, MegaDepth and YFCC first from [driven](https://drive.google.com/drive/folders/1TE_zJlKfPFRLeIrtq5iMBBjg-XaovNon).
28 | 
29 | For the data utilized for training, we use the same training data as [LoFTR](https://github.com/zju3dv/LoFTR) does.
30 | 
31 | 
32 | ### Megadepth validation
33 | For different scales, you need edit [megadepth_test_scale_1000](configs/data/megadepth_test_scale_1000.py).
34 | 
35 | ```shell
36 | # with shell script
37 | bash ./scripts/reproduce_test/outdoor_ada_scale.sh
38 | ```
39 | 
40 | 
41 | ### Reproduce the testing results for yfcc datasets
42 | ```shell
43 | # with shell script
44 | bash ./scripts/reproduce_test/yfcc100m.sh
45 | ```
46 | 
47 | <br/>
48 | 
49 | 
50 | ## Training
51 | We train AdaMatcher on the MegaDepth datasets following [LoFTR](https://github.com/zju3dv/LoFTR/blob/master/docs/TRAINING.md). And the results can be reproduced when training with 32gpus. Please run the following commands:
52 | 
53 | ```
54 | sh scripts/reproduce_train/outdoor_ada.sh
55 | ```
56 | ## Acknowledgement
57 | 
58 | This repository is developed from `LoFTR`, and we are grateful to its authors for their [implementation](https://github.com/zju3dv/LoFTR).
59 | 
60 | ## Citation
61 | 
62 | If you find this code useful for your research, please use the following BibTeX entry.
63 | 
64 | ```bibtex
65 | @article{Huang2023adamatcher,
66 |   title={Adaptive Assignment for Geometry Aware Local Feature Matching},
67 |   author={Dihe Huang, Ying Chen, Yong Liu, Jianlin Liu, Shang Xu, Wenlong Wu, Yikang Ding, Fan Tang, Chengjie Wang},
68 |   journal={{CVPR}},
69 |   year={2023}
70 | }
71 | ```
72 | 


--------------------------------------------------------------------------------
/assets/network.pdf:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/AbyssGaze/AdaMatcher/19cb67cb76586da484d931d3d346a258224ac976/assets/network.pdf


--------------------------------------------------------------------------------
/assets/network.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/AbyssGaze/AdaMatcher/19cb67cb76586da484d931d3d346a258224ac976/assets/network.png


--------------------------------------------------------------------------------