This is the implementation of the paper:
I. Rocco, R. Arandjelović and J. Sivic. Convolutional neural network architecture for geometric matching. [website][CVPR version][Extended TPAMI version]
See requirements.txt
Please see the demo.py
script or the demo_notebook.ipynb
Jupyter Notebook.
You can train the model using the train.py
script in the following way:
python train.py --geometric-model affine
For a full set of options, run python train.py -h
.
- For now it is implemented to log on TensorBoard just scalars of train and val loss
- It is possible to specify a --logdir as a parameter, otherwise the logging folder will be named as the checkpoint one with _tb_logs as suffix
- N.B. If is intended to use as logdir a GCP bucket it is necessary to install Tensorflow
You can evaluate the trained models using the eval.py
script in the following way:
python eval.py --model-1 trained_models/best_streetview_checkpoint_adam_hom_grid_loss_PAMI.pth.tar --eval-dataset pf
You can also evaluate a two-stage model in the following way:
python eval.py --model-1 trained_models/best_streetview_checkpoint_adam_hom_grid_loss_PAMI.pth.tar --model-2 trained_models/best_streetview_checkpoint_adam_tps_grid_loss_PAMI.pth.tar --eval-dataset pf
The eval.py
scripts implements the evaluation on the PF-Willow/PF-PASCAL/Caltech-101 and TSS datasets. For a full set of options, run python eval.py -h
.
Model | PF-Willow (PCK) |
---|---|
[Affine - VGG - StreetView] | 48.4 |
[Homography - VGG - StreetView] | 48.6 |
[TPS - VGG - StreetView] | 53.8 |
If you use this code in your project, please cite us using:
@InProceedings{Rocco17,
author = {Rocco, I. and Arandjelovi\'c, R. and Sivic, J.},
title = {Convolutional neural network architecture for geometric matching},
booktitle = {{Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition}},
year = {2017},
}
or
@Article{Rocco18,
author = {Rocco, I. and Arandjelovi\'c, R. and Sivic, J.},
title = {Convolutional neural network architecture for geometric matching},
journal = {{IEEE Transactions on Pattern Analysis and Machine Intelligence}},
number = {41},
pages = {2553--2567},
year = {2018},
}