line-shaped artifacts in reconstructed scenes

[AnarchistKnight]

Good day.

I am trying to reconstruct and render scemes in AriaSynthesisDataset. The first image below is one that uses the complete sequence. Considering some sequences contain over 1500 frames, I sample every 3 frames to reduce time consumption. The second image below is the result. The line-shaped artifacts of the floor are more obvious than the one that builds from all the frames. I guess, building from more frames in some degree averages out the difference in brightness. I removed vignetting (vignetting mask here) before reconstructing. Unfortunately, subsequent frames still differ in brightness. Any suggestion to remove or reduce the artifacts?

[suvampatra]

Please have a look at the discussion here #99

and this pull request which has a demo #125

[AnarchistKnight]

Please have a look at the discussion here #99

and this pull request which has a demo #125

Do you think I use the wrong vignetting mask?

[Rock-100]

@AnarchistKnight I think the masks here (https://github.com/facebookresearch/projectaria_tools/tree/main/data/vignetting_mask) are not for ASE dataset, as we can see their sizes are not the same with images in ASE. @suvampatra Could you please tell me how to find a proper mask?

[suvampatra]

The idea behind using the vignetting mask is things outside the vignette mask is not to be used. To my knowledge, the best way of inverting is by using the method described by @captain-sysadmin in #99

[ChenYutongTHU]

@AnarchistKnight I think the masks here (https://github.com/facebookresearch/projectaria_tools/tree/main/data/vignetting_mask) are not for ASE dataset, as we can see their sizes are not the same with images in ASE. @suvampatra Could you please tell me how to find a proper mask?

I agree with this. The image size in ASE is (704,704). However only vignette masks of (1408,1408) and (2816,2816) are provided.

Where can we find the vignette masks for ASE?

Thanks