search

gmberton / image-matching-models

Code to easily try 30 (and growing) different image matching methods
https://earthloc-and-earthmatch.github.io/
BSD 3-Clause "New" or "Revised" License
247 stars 17 forks source link

Can i Use these for 360 panorama images #9

Open paidiakileswar opened 3 weeks ago

paidiakileswar commented 3 weeks ago

Hai , Great Work

Can these models works fine for 360 images which are captured by insta 360 cameras - basically have some distortion in them with difference in angles and views ;

How well did matching algorithms works in these images ! Can i know

Thank you

gmberton commented 3 weeks ago

Hi, we never tried on 360 images. I doubt any of these methods have been trained on 360 images, but the keypoints extraction should give acceptable results, and perhaps even the matching. RANSAC post processing would fail (which is the output that you get as inliers_kpts). But trying most models would take you a few seconds so feel free to try and do let us know how it goes :)

paidiakileswar commented 2 weeks ago

Thanks for Quick Reply @gmberton -> I am going to try that on 360 images , Asap my testing is done , Here I will do share my results ,

A small help - Can you please share any Graph or Benchmarks that you have tested on normal images , To get to know - How did each algorithm worked , please share here that will be useful for my comparison

alexstoken commented 2 weeks ago

Are your views similar to those addressed in SphereGlue? That might have a hint as to how to proceed if so.

Re: benchmarks, there are a few that come to mind with image matching. Most are reported in the papers associated with the models (see table in README). This is also a great paper that is a few years old now but discusses many aspects of matching. Newer papers will usually contain tables comparing their results to other recent works.

Curious to hear how your experiments go!

gmberton commented 2 weeks ago

@paidiakileswar if you want to have a look at existing benchmarks using all these models you can check Mismatched: Evaluating the Limits of Image Matching Approaches and Benchmarks or our EarthMatch paper (only on satellite images). Looking forward to seeing your results!