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
1.34k stars 149 forks source link

Query regarding data loading for custom dataset #113

Open sumitanilkshirsagar opened 4 weeks ago

sumitanilkshirsagar commented 4 weeks ago

Hi, I am working on a underwater 3D reconstruction project, and wanted to first establish a backbone 3D reconstruction library and then build on top of it. HengyiWang(Co-SLAM ) suggested thaT your code would work better for under water 3D purposes. I wanted to check how I can build a dataset and config, I am using a stereo camera (ZED camera from stereolabs) with a resolution of HD2K ( 2208x1242)

What I tried so far: 1) I tried to resize my rgb,depth images to 640x480 and placing them in the TUM dataset folder.

2)Tried running 2208x1242 and changing resolution in config file.

sumitanilkshirsagar commented 4 weeks ago

1) I tried to resize my rgb,depth images to 640x480 and placing them in the TUM dataset folder. Error:

SplaTAM/scripts/splatam.py", line 107, in get_pointcloud point_cld = torch.cat((pts, cols), -1) RuntimeError: Sizes of tensors must match except in dimension 1. Expected size 307200 but got size 409600 for tensor number 1 in the list.

Can you help me with what these tensors denote, and where is mismatch coming from?

Nik-V9 commented 2 weeks ago

Hi, Thanks for your interest in our work!

You will need to implement your own dataloader using the ones present here as reference: https://github.com/spla-tam/SplaTAM/tree/main/datasets/gradslam_datasets

You will also need a corresponding config file for the data, which contains the intrinsic & depth PNG scale information (pixel value to depth mapping): https://github.com/spla-tam/SplaTAM/tree/main/configs/data

Regarding the issue you are facing, there might be a mismatch between your RGB & depth image size.