ufukefe / DFM

Python (Pytorch) and Matlab (MatConvNet) implementations of CVPR 2021 Image Matching Workshop paper DFM: A Performance Baseline for Deep Feature Matching
Other
197 stars 22 forks source link

Adaptable to match a object in changeable scales? #11

Open miaTian99 opened 1 year ago

miaTian99 commented 1 year ago

Hi, the non-training trick is quite rewarding, and I appreciate the outstanding work of the DFM.

There are some questions confuse me for a period of time:

  1. Whether DFM adapts to matching object in changeable scales or not ? Apart from illuminations and viewpoints, the scale of a object is changing all the time. According to my experiments, the DFM always fails when drone images and satellite images have huge differences in scale.
  2. Can problem 1 be solved by change a pretrained backbone model ? I have tried to improve the VGG with SPP, but this didn't seem to work...I have no idea about whether this kind of test is effective with the DFM.

Could you please give me some suggestions about that?
thx a lot !!! @ufukefe @kutalmisince