Zixin Luo

The training code is under preparation for both geodesc and contextdesc (a follow-up work of geodesc). Since the projects are industry-driven, it may take some time to resolve the patent...

Sorry for taking forever to resolve the license issue. The training code is now publicly available, together with two follow-up works published in CVPR19 and CVPR20. Please visit [TFMatch](https://github.com/lzx551402/tfmatch) for...

Hi, Keypoint sampling regarding its feature size (for a DoG-like detector) is a common practice in SfM pipelines (visualSFM, COLMAP etc.), which is the target application for this work. Usually,...

Hi, We used the ground-truth correspondences as provided in [GL3D](https://github.com/lzx551402/GL3D). We also provided some helper function to help to parse the data.

Sorry for the late response. Basically, σ denotes the feature scale, which can be obtained from a keypoint detector that constructs image pyramid, such as SIFT. The size of a...

I also write an example to address this issue. Please refer to this [discussion](https://github.com/lzx551402/GL3D/issues/6).

Sorry for taking forever to resolve the license issue. We have now made this project publicly available, please visit [TFMatch](https://github.com/lzx551402/tfmatch) if you are still interested.

Hi, `cameras.txt` records the original camera parameters, including image size. To obtain camera parameters, such as focal length of downscaled images, you may first compute the rescaling ratio in x...

Sorry for taking forever to resolve the license issue. We have now made this project publicly available, please visit [TFMatch](https://github.com/lzx551402/tfmatch) if you are still interested.

1. Thank you for pointing out the wrong dependencies, and I have updated ``requirements`` accordingly. 2. Those warning messages raised from tensorflow2.0 would not affect this program. 3. Did you...