search

spla-tam / SplaTAM

SplaTAM: Splat, Track & Map 3D Gaussians for Dense RGB-D SLAM (CVPR 2024)
https://spla-tam.github.io/
BSD 3-Clause "New" or "Revised" License
1.58k stars 174 forks source link

Query regarding first camera pose #111

Closed suvzzc closed 5 months ago

suvzzc commented 6 months ago

Thank you for your excellent work. I would like to ask why the pose read from the dataset "groundtruth. txt" is used? Isn't it position estimation? Why is the pose from the dataset used? 1

Nik-V9 commented 5 months ago

Hi, Thanks for your interest!

We configure our dataset to return poses relative to the first frame. Hence, the first camera pose that we load with dataset[0] is always identity: https://github.com/spla-tam/SplaTAM/blob/1a5d407c3a145f69691d49267411cb65e5d5c992/scripts/splatam.py#L529

However, when you want to use absolute poses for SLAM (which is conventionally not preferred and done), the code using dataset[0] supports this (as opposed to just hardcoding the first pose to identity).

Also, these additional details about pose estimation might be relevant:

  1. https://github.com/spla-tam/SplaTAM/issues/30#issuecomment-1853483102
  2. https://github.com/spla-tam/SplaTAM/issues/28
suvzzc commented 5 months ago

Hi, Thanks for your interest!

We configure our dataset to return poses relative to the first frame. Hence, the first camera pose that we load with dataset[0] is always identity:

https://github.com/spla-tam/SplaTAM/blob/1a5d407c3a145f69691d49267411cb65e5d5c992/scripts/splatam.py#L529

However, when you want to use absolute poses for SLAM (which is conventionally not preferred and done), the code using dataset[0] supports this (as opposed to just hardcoding the first pose to identity).

Also, these additional details about pose estimation might be relevant:

1. [a small question about ref_w2c #30 (comment)](https://github.com/spla-tam/SplaTAM/issues/30#issuecomment-1853483102)

2. [Q about pose estimation #28](https://github.com/spla-tam/SplaTAM/issues/28)

Thank you for your reply. I re-read the code and realized that I misunderstood. Here gt_pose uses relative pose and it is not used in the trace section.