two questions about ratio loss and training patch extraction

vbalnt / tfeat

TFeat descriptor models for BMVC 2016 paper "Learning local feature descriptors with triplets and shallow convolutional neural networks"

MIT License

148 stars 45 forks source link

two questions about ratio loss and training patch extraction #10

Closed RuibinMa closed 6 years ago

RuibinMa commented 6 years ago

Hi, thank you for the great work first. May I ask two questions: 1) Basically ratio loss is unsuitable (doesn't converge at all) for keypoint matching according to your paper, so the best choice is ranking with anchor swap. Am I correct? 2) How are the training patches from Lowe's DoG keypoints extracted? Is it just a fixed-size window cropping or based on the keypoints' scale ( possibly orientation as well?) followed by a normalization step?

Thanks in advance.

RuibinMa commented 6 years ago

I'm also wondering how is the network's performance on rotating and scale changing scenarios.

vbalnt commented 6 years ago

@RuibinMa please check the test code, where you can find the runtime example using openCV.