No good result on new test video about skyar HOT 13 CLOSED

Hello! First of all, nice work!
What do you mean by "rich sky texturing"? Can you explain a bit and give an example?
Thank you!

Thanks for your comments! I can give you an example. In the figure below, the left one is with "rich sky textures" but the right one is not.

from skyar.

Hi, thanks for your feedback. I watched your video and I believe the problem lies in the sky background. As mentioned in my preprint paper, when there are no textures in the sky, the motion of the sky background cannot be accurately modeled. This is one of the limitations of the method.
I have updated the skyboxengine.py, where this time I limited the motion to translation + rotation, and raise the threshold of the number of effective matching points (3 ->10). You can try again and tell me whether it works this time. With any luck, this update will work on your case. But as I mentioned above, the most reliable solution is to choose a video with a rich sky texture for testing. Good luck.

from skyar.

Hello! First of all, nice work!
What do you mean by "rich sky texturing"? Can you explain a bit and give an example?
Thank you!

from skyar.

Thank you, I understand now! This problem occurs because of the dataset? I mean, less images with clear skies day/night?

from skyar.

Thank you, I understand now! This problem occurs because of the dataset? I mean, less images with clear skies day/night?

I believe the problem is in the motion estimation rather than the dataset training. If you set "save_jpgs" in the config file to "true" and check those sky region detection results, you will see those images with clear sky regions are nicely detected. However, when there are not enough feature points in the sky, the tracker may fail to find any effective matching points between adjacent frames, and that why we need a rich textured sky to make sure the motion of the virtual background is accurately modeled.

from skyar.

I understand! Thank you very much for the info! Once again, nice work :)

from skyar.

I understand! Thank you very much for the info! Once again, nice work :)

Cheers!

from skyar.

Hi, thanks for your feedback. I watched your video and I believe the problem lies in the sky background. As mentioned in my preprint paper, when there are no textures in the sky, the motion of the sky background cannot be accurately modeled. This is one of the limitations of the method.
I have updated the skyboxengine.py, where this time I limited the motion to translation + rotation, and raise the threshold of the number of effective matching points (3 ->10). You can try again and tell me whether it works this time. With any luck, this update will work on your case. But as I mentioned above, the most reliable solution is to choose a video with a rich sky texture for testing. Good luck.

Will try the fixes, thx a lot.

from skyar.

@jiupinjia I made a new test video with the settings. I think it´s a little bit better but still have the jumping effects.

Here is the new video:
https://www.youtube.com/watch?v=rzJWS3jqiy8

Maybe we work with pre-marker in the sky that can help to match the points for tracking?

from skyar.

@jiupinjia I made a new test video with the settings. I think it´s a little bit better but still have the jumping effects.

Here is the new video:
https://www.youtube.com/watch?v=rzJWS3jqiy8

Maybe we work with pre-marker in the sky that can help to match the points for tracking?

Seems the problem is still there. Can you show me the input video?

from skyar.

@jiupinjia Sure, you can download it here, I remove the zip in 48 hours.
http://rebotnix.com/tmp/sky_videos.zip

Thanks for your help. Will test any changes.

from skyar.

@jiupinjia Sure, you can download it here, I remove the zip in 48 hours.
http://rebotnix.com/tmp/sky_videos.zip

Thanks for your help. Will test any changes.

Hi, I have carefully checked your input videos, and I believe the problem is in the sky background. As I mentioned in my preprint paper Sec 4.4, one of the limitations of the method is that it only works with input videos with rich sky textures. When there are no textures, the motion of the virtual camera cannot be accurately modeled. To better understand this, I can give you an example. In the figure below, the left one shows a frame with rich sky textures but the right one does not.

I have put SkyAR version 2.0 on my agenda, where I will be working on a more robust motion estimation (motion propagation from foreground to background) and hopefully, the SkyAR 2.0 will have no more requirements on the sky textures. Thanks again for your valuable feedback.

from skyar.

I fully understand it and i thank you that you take your time to check the videos. It´s a great project and I will follow it for sure. Good luck.

from skyar.