mega-nerf icon indicating copy to clipboard operation
mega-nerf copied to clipboard

Different betweem with the offical dataset Sci-Art and mine

Open chufall opened this issue 1 year ago • 4 comments

Hi , I still confused with building my own dataset following the method in Readme and discussions in https://github.com/cmusatyalab/mega-nerf/issues/6 .

In order to verfy whether i'm right, I built the dataset with Sci-Art photos and compared with the same set provided in https://storage.cmusatyalab.org/mega-nerf-data/sci-art-pixsfm.tgz

If I'm right , the visualization result should be the same. Here are the results. 1. official

74d1bb021ca01c2bedb4248067a4221

2. mine 9d8f686c363f7800a9215b9c844cf94

You will find that the cameras in official set are aligned the X axis(red) , but mines are aligned the Z axis(blue). So I think there is a step before using the convert script, or there're anything wrong using the pixsfm。 How to do it? Thank you very much!

Sincerely. QC

chufall avatar Jul 10 '23 03:07 chufall

hello, I'm also trying to eval sci-art dataset, but there are 3 authors' datasets on https://vcc.tech/UrbanScene3D and each has 3 zips. So which one is the final choice? Could you share your choice? Thanks a lot!

A-pril avatar Aug 16 '23 08:08 A-pril

hello, I'm also trying to eval sci-art dataset, but there are 3 authors' datasets on https://vcc.tech/UrbanScene3D and each has 3 zips. So which one is the final choice? Could you share your choice? Thanks a lot!

https://github.com/Linxius/UrbanScene3D#urbanscene3d-v1
hi, use this link to download

chufall avatar Aug 29 '23 06:08 chufall

Hi , I still confused with building my own dataset following the method in Readme and discussions in #6 .

In order to verfy whether i'm right, I built the dataset with Sci-Art photos and compared with the same set provided in https://storage.cmusatyalab.org/mega-nerf-data/sci-art-pixsfm.tgz

If I'm right , the visualization result should be the same. Here are the results. 1. official

74d1bb021ca01c2bedb4248067a4221

2. mine 9d8f686c363f7800a9215b9c844cf94

You will find that the cameras in official set are aligned the X axis(red) , but mines are aligned the Z axis(blue). So I think there is a step before using the convert script, or there're anything wrong using the pixsfm。 How to do it? Thank you very much!

Sincerely. QC

Hello, have you successfully customized the dataset and trained it. Can you provide your method

lhc991025 avatar Dec 19 '23 05:12 lhc991025

@lhc991025 I also need to customized the dataset and trained it, maybe we can talk about it. my weixin is d19965161832

AI-slam avatar Dec 19 '23 06:12 AI-slam