search

NVlabs / BundleSDF

[CVPR 2023] BundleSDF: Neural 6-DoF Tracking and 3D Reconstruction of Unknown Objects
https://bundlesdf.github.io/
Other
992 stars 109 forks source link

connection reset by peer #153

Closed zahid990170 closed 2 months ago

zahid990170 commented 4 months ago

Hi,

I tried to run bundleSDF using the two datasets provided, i.e., milk dataset, and HO3D dataset. For the milk dataset the following command was issued:

python run_custom.py --mode run_video --video_dir 2022-11-18-15-10-24_milk --out_folder bundlesdf_2022-11-18-15-10-24_milk --use_segmenter 1 --use_gui 0 --debug_level 2

For HO3D dataset the following command was issued:

python run_ho3d.py --video_dirs /mnt/9a72c439-d0a7-45e8-8d20-d7a235d02763/DATASET/HO3D_v3/evaluation/SM1 --out_dir /home/bowen/debug/ho3d_ours

In both cases, the program runs for a while, but stops with the following message

connection reset by peer It was preceded by "RuntimeError: Unable to find a valid cuDNN algorithm to run convolution when training on host machine with gpu"

Is the above error because of lower specs of machine.

My GPU model:

image

VRAM: 122MiB / 7973MiB ~ 8 GB

pytorch 1.11 cuda 11.3

What are reasonable specification for GPU model and memory that can make the program run to completion.

thanks,

wenbowen123 commented 4 months ago

https://github.com/NVlabs/BundleSDF/issues/151 seems related