cvg / Hierarchical-Localization

Visual localization made easy with hloc
Apache License 2.0
3.2k stars 593 forks source link

using hloc for reconstruct with existed matches #269

Open ronhaim311 opened 1 year ago

ronhaim311 commented 1 year ago

Hello, if I already have images and matches between them, can I use the hloc to build and reconstruct 3D model? if I can so how? I saw it has a stage for feature extraction, but I have already the matches ready for reconstruction, can I skip this stage?

sarlinpe commented 1 year ago

Sorry for the late reply. You'd need to write keypoints to a feature h5 file and matches between them to another h5 files. Then you can call hloc.reconstruction. See the loading functions for more details on the format:

features:

{
  image_name: {
    'keypoints': array with shape Nx2 
  },
}

matches:

{
  image_name_1/image_name_2: {
    'matches': array with shape N, where N is the number of keypoints in image 1
  },
}
Tiangj9178 commented 1 day ago

很抱歉回复晚了。您需要将关键点写入特征 h5 文件,并将它们之间的匹配项写入另一个 h5 文件。然后你可以调用 hloc.reconstruction。有关格式的更多详细信息,请参阅 loading functions

特征:

{
  image_name: {
    'keypoints': array with shape Nx2 
  },
}

比赛:

{
  image_name_1/image_name_2: {
    'matches': array with shape N, where N is the number of keypoints in image 1
  },
}

Hello! I am also interested in this question. How can I read the feature points from the database.db file of colmap and save them in two .h5 files with the feature matching, and ensure that the format is consistent with the matches.h5 file obtained by running in hloc?