search

cvg / pixloc

Back to the Feature: Learning Robust Camera Localization from Pixels to Pose (CVPR 2021)
Apache License 2.0
742 stars 93 forks source link

With custom data #41

Open CuiYan27 opened 2 years ago

CuiYan27 commented 2 years ago

Hi, thanks for your great work! I want to use pixloc to implement a localiazation task.

I followed these steps:

  1. I create my own datasets that I capture by my own camera.
  2. I use netvlad to create global descriptors.
  3. I use superpoint to create local descriptors.
  4. I use retrieval to create pairs.
  5. Match the pairs use superglue.
  6. Follow the reconstruction to create a model--sfm_superpoint+superglue
  7. Finally, I use the model to localize query image.

Unfortunately, I can't achieve a precise result with pixloc. The SfM model looks well when I visualized in the COLMAP GUI and the estimated camera poses roughly correct. What should I do to improve it?

tieguanyin803 commented 1 year ago

I have the same question.

tsattler commented 1 year ago

pixloc does camera pose refinement starting from an initial pose estimate. If the initial pose estimate is too far off from the true pose, pixloc can fall into local minima, preventing it from recovering the camera pose accurately.

tworkool commented 4 months ago

@CuiYan27 Could you please provide a code example? I am also trying to use PixLoc on a custom dataset but I have multiple problems... I use HLoc to do exactly what you described.