README.md

[AAAI 2023] Semi-supervised Deep Large-baseline Homography Estimation with Progressive Equivalence Constraint. [Paper].

Hai Jiang^1,2, Haipeng Li^2,3, Yuhang Lu⁴, Songchen Han¹, Shuaicheng Liu^2,3

1.Sichuan University, 2.University of Electronic Science and Technology of China,

3.Megvii Technology, 4.University of South Carolina

Presentation video:

[Youtube] and [Bilibili]

Pipeline

Dependencies

pip install -r requirements.txt

- This repo includes GOCor as git submodule. You need to pull submodules with

git submodule update --init --recursive
git submodule update --recursive --remote

Download the raw dataset

Please refer to Content-Aware Unsupervised Deep Homography Estimation..

- Dataset download links: [GoogleDriver], [BaiduYun] (key:gvor)

- Unzip the data to the directory "./dataset"

- Run "video2img.py"

Be sure to scale the image to (640, 360) since the point coordinate system is based on the (640, 360).
e.g. img = cv2.imresize(img, (640, 360))

- Using the images in "train.txt" and "test.txt" for training and evaluation, the manually labeled evaluation files can be downloaded from: [GoogleDriver], [BaiduYun](key:i721).

Pre-trained model

The model provided below is the retrained version(with some differences in reported results)

model	RE	LT	LL	LF	SF	Avg	Model
Pre-trained	1.66	5.49	4.11	7.57	6.95	5.16	[Baidu](code: 0jfu) [Google]

How to train?

You need to modify dataset/data_loader.py slightly for your environment, and then

python train.py --model_dir experiments/base_model/ 

How to test?

python evaluate.py --model_dir experiments/base_model/ --restore_file xxx.pth

Citation

If you use this code or ideas from the paper for your research, please cite our paper:

@InProceedings{jiang_2023_aaai,
    author  = {Jiang, Hai and Li, Haipeng and Lu, Yuhang and Han, Songchen and Liu, Shuaicheng},
    title = {Semi-supervised Deep Large-baseline Homography Estimation with Progressive Equivalence Constraint}},
    booktitle = {Proceedings of the Thirty-Seventh AAAI Conference on Artificial Intelligence (AAAI)}
    year = {2023}
}