CUDA out-of-memory while running run.py with Bonn dataset, Wild-SLAM Mocap Dataset

[Fnhid]

Thank you for your awesome work.

I tried running your run.py with the Bonn Dataset and Wild-SLAM Mocap Dataset.

However, it fails with a CUDA out-of-memory error during the TRACKER running

Traceback (most recent call last):
  File ".../multiprocessing/process.py", line 108, in run
    self._target(*self._args, **self._kwargs)
  File "/root/WildGS-SLAM/src/slam.py", line 125, in mapping
    self.mapper.run()
  File "/root/WildGS-SLAM/src/mapper.py", line 180, in run
    self.initialize_mapper(video_idx)
  File "/root/WildGS-SLAM/src/mapper.py", line 779, in initialize_mapper
    self.gaussians.extend_from_pcd_seq(
  File "/root/WildGS-SLAM/thirdparty/gaussian_splatting/scene/gaussian_model.py", line 265, in extend_from_pcd_seq
    self.create_pcd_from_image(cam_info, init, scale=scale, depthmap=depthmap)
  File "/root/WildGS-SLAM/thirdparty/gaussian_splatting/scene/gaussian_model.py", line 136, in create_pcd_from_image
    return self.create_pcd_from_image_and_depth(cam, rgb, depth, init)
  File "/root/WildGS-SLAM/thirdparty/gaussian_splatting/scene/gaussian_model.py", line 203, in create_pcd_from_image_and_depth
    distCUDA2(torch.from_numpy(np.asarray(pcd.points)).float().cuda()),
MemoryError: std::bad_alloc: cudaErrorMemoryAllocation: out of memory

My setup is:

- GPU: Tesla V100 32GB 
- OS: Ubuntu 22.04 (Docker)
- Python: 3.10.19 (using conda)
- PyTorch: 2.1.0+cu118
- MMVC: 1.7.2

I also lowered the image output resolution as suggested in the README. However, the error persists.
Could you provide any recommended solutions or debugging approaches?

[Jianhao-zheng]

Hi, thanks for your interest in our work. I can run wildgs-slam on bonn and wild-slam dataset with a 4090 (24GB). Can you check the GPU usage of your computer before running wildgs-slam? Maybe some processes have already occupied memory.