Open Neural Network Exchange (ONNX) compatible implementation of XFeat: Accelerated Features for Lightweight Image Matching. The ONNX model format allows for interoperability across different platforms with support for multiple execution providers, and removes Python-specific dependencies such as PyTorch. Source code in C++ is also provided to test the model!.
TODO or Pip install something.
TODO
As the ONNX Runtime has limited support for features like dynamic control flow, certain configurations of the models cannot be exported to ONNX easily. These caveats are outlined below.
- Only batch size
1is currently supported. This limitation stems from the fact that different images in the same batch can have varying numbers of keypoints, leading to non-uniform (a.k.a. ragged) tensors. For this reason the code differ from the original project: batch operations have been removed.
- Actually Dense Multiscale model is not supported. Possible Future Work!
- To compare the Pytorch/Onnx/C++ models, the images in the assets folder were used. It can be seen in the results that the Python Pytorch/ONNX results are very similar to each other. For the C++ part, the results may vary slightly, particularly with a noticeable upward sliding of points compared to the Python versions. However, when using the images without applying padding (image_800x608), the sliding does not occur. Could some scaling/rounding factor be getting lost somewhere?
Project taken from: XFeat
Please cite the paper:
@INPROCEEDINGS{potje2024cvpr,
author={Guilherme {Potje} and Felipe {Cadar} and Andre {Araujo} and Renato {Martins} and Erickson R. {Nascimento}},
booktitle={2024 IEEE / CVF Computer Vision and Pattern Recognition (CVPR)},
title={XFeat: Accelerated Features for Lightweight Image Matching},
year={2024}}Ty for the amazing job! Guilherme Potje · Felipe Cadar · Andre Araujo · Renato Martins · Erickson R. Nascimento