Skip to content

Latest commit

 

History

History

COMMON

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
 
 
 
 
 
 

COMMON

Official implementation of the following paper:

  • Yijie Lin, Mouxing Yang, Jun Yu, Peng Hu, Changqing Zhang, Xi Peng. "Graph Matching with Bi-level Noisy Correspondence". ICCV, 2023. [paper]

COMMON proposes a deep graph matching pipeline to solve Noisy Correspondence in GM problem, which refers to node-level noisy correspondence (NNC) and edge-level noisy correspondence (ENC).

COMMON, ICCV 2023

As shown above, due to the poor recognizability and viewpoint differences between images, it is inevitable to inaccurately annotate some keypoints with offset and confusion, leading to the mismatch between two associated nodes (NNC). The noisy node-to-node correspondence will further contaminate the edge-to-edge correspondence (ENC).

It proposes to combine the following components to formulate the GM pipeline:

  • VGG16 CNN to extract image features
  • SplineCNN to embed the graph information
  • GM customized quadratic contrastive learning
  • Momentum teacher to adaptively penalizing the noisy assignments

Benchmark Results

PascalVOC - 2GM

experiment config: experiments/vgg16_common_voc.yaml

pretrained model: google drive

model year aero bike bird boat bottle bus car cat chair cow table dog horse mbkie person plant sheep sofa train tv mean
COMMON 2023 0.6560 0.7520 0.8080 0.7950 0.8930 0.9230 0.9010 0.8180 0.6160 0.8070 0.9500 0.8200 0.8160 0.7950 0.6660 0.9890 0.7890 0.8090 0.9930 0.9380 0.8270

Willow Object Class - 2GM

experiment config: experiments/vgg16_common_willow.yaml

pretrained model: google drive

model year remark Car Duck Face Motorbike Winebottle mean
COMMON 2023 - 0.9760 0.9820 1.0000 1.0000 0.9960 0.9910

SPair-71k - 2GM

experiment config: experiments/vgg16_common_spair71k.yaml

pretrained model: google drive

model year aero bike bird boat bottle bus car cat chair cow dog horse mtbike person plant sheep train tv mean
COMMON 2023 0.7730 0.6820 0.9200 0.7950 0.7040 0.9750 0.9160 0.8250 0.7220 0.8800 0.8000 0.7410 0.8340 0.8280 0.9990 0.8440 0.9820 0.9980 0.8450

File Organization

├── model.py
|   the implementation of training/evaluation procedures of COMMON
├── model_config.py
|   the declaration of model hyperparameters
└── sconv_archs.py
    the implementation of spline convolution (SpilneCNN) operations, the same with BBGM

Note that the network structure of projection layer is defined in model.py, which is a two-layer MLP with ReLU activation.

Credits and Citation

Please cite the following paper if you use this model in your research:

@article{lin2023graph,
  title={Graph Matching with Bi-level Noisy Correspondence},
  author={Lin, Yijie and Yang, Mouxing and Yu, Jun and Hu, Peng and Zhang, Changqing and Peng, Xi},
  journal={IEEE International Conference on Computer Vision},
  year={2023}
}