search

cvg / LightGlue

LightGlue: Local Feature Matching at Light Speed (ICCV 2023)
Apache License 2.0
3.15k stars 291 forks source link

sift+lightglue #122

Open LXXIANG12 opened 2 months ago

LXXIANG12 commented 2 months ago

sift描述符是128维，lightglue描述符是256维，我该怎么使用SIFT+lightglue呢？

pxy522 commented 2 months ago

好像训练的时候可以改

LXXIANG12 commented 2 months ago

希望训练的时候可以修改

官方公布了SIFT+lightglue的预训练权重，我想先用这个模型试试看效果，但是使用过程报错了 image 以下是我的示例： image 我第一次使用不知道改哪里

pxy522 commented 2 months ago

image zh 这？

LXXIANG12 commented 2 months ago

抱歉，改这里好像不行 image

Phil26AT commented 2 months ago

If I understand your issue correct (please post your issues in english), you want to use SIFT with LightGlue. You need to load the correct weights in LightGlue, with:

extractor = SIFT(max_num_keypoints=None).eval().cuda()  # load the extractor
matcher = LightGlue(features='sift').eval().cuda()  # load the matcher

Note that you need to specify which weights to load in the constructor of LightGlue (the SIFT setup is defined here).