search

cvg / pixloc

Back to the Feature: Learning Robust Camera Localization from Pixels to Pose (CVPR 2021)
Apache License 2.0
735 stars 92 forks source link

Question: TypeError #37

Closed CuiYan27 closed 2 years ago

CuiYan27 commented 2 years ago

Hi, Thanks for your amazing work. I encountered the following error when I tried to run code with a weak textured indoor scene dataset.

In the following case, is the value of this weight related to the quality of the dataset, such as the texture of the scene?

image image image

sarlinpe commented 2 years ago

These global descriptors are incorrect. They should be N-dimensional normalized vectors. Here they have values [1920, 1080] or [3840, 2160] - is this the sizes of your images? How did you compute the descriptors?

CuiYan27 commented 2 years ago

Thanks for your reply! The values of [1920, 1080] and [3840, 2160] are sizes of my input image.

I calculate the global descriptors using the hloc tools. The command is: python -m hloc.extract_features --image_dir datasets/Scenes/scene5/photos/all --export_dir outputs/Scenes/scene5 --conf netvlad

sarlinpe commented 2 years ago

Can you please try out PR https://github.com/cvg/pixloc/pull/39 and let me know if that solves the issue?

CuiYan27 commented 2 years ago

With your help #39 , I have solved the issue successfully! Thank you very much!