problem in matching

[BlingHe]

Hi Professor，thank you for sharing the code!

I have extracted features by using the pretrained model and matched them following this issue ，but the matching result seems not good enough, even when i matched the picture with its rotated one, there also are a lot of error pairs in result picture. Is there something i can do to promote the matching performance? May i need to retrain the model?

I am looking forward to your reply, thanks a lot!

here is the result when the picture were matched with its rotated one.

[kmyi]

Hi, which pretrained model are you using? In the TF-LIFT repo, there's a rotation invariant one. Maybe try that one? We also have lf-net which should also be better.

[Jinniuzuo0516]

@BlingHe hello, may i ask how run the pre-training model successfully?