search

ubc-vision / image-matching-benchmark

Public release of the Image Matching Benchmark: https://image-matching-challenge.github.io
https://image-matching-challenge.github.io
Apache License 2.0
527 stars 93 forks source link

Generating the SfM bags for new scenes #41

Closed sarlinpe closed 3 years ago

sarlinpe commented 3 years ago

I want to do some additional validation of the SfM task using training scenes of the Phototourism dataset. How should I generate the bags? The paper says that the bags are randomly sampled from the original set of 100 images per scene – with a co-visibility check (Section 4.5). I tried to:

  1. pick subsets where all images share 3D points: this is too restrictive and yields very few to no bags of 25 images.
  2. perform random walks on the visibility graph with a minimum ratio of covisible 3D points: this yields chains that are more difficult to reconstruct than the bags in the validation scenes.

Any pointer or code would be much appreciated, thanks!

etrulls commented 3 years ago

It was pretty simple, from what I recall. We refactored the code and there's a new script to do this, but I can't find it right now. @jyh2005xx can you please point Paul-Edouard to it? Thanks.

sarlinpe commented 3 years ago

@jyh2005xx helped out, thanks!

biggiantpigeon commented 1 year ago

@jyh2005xx helped out, thanks!

Hi, I'm trying to generating my own test set. Could you please show me the script you use? I am also wondering how to generate the folders like 'calibration', 'depth_maps' in a scene, could you give me some hint? Thanks a lot!