search

Shuijing725 / CrowdNav_Prediction_AttnGraph

[ICRA 2023] Intention Aware Robot Crowd Navigation with Attention-Based Interaction Graph
https://sites.google.com/view/intention-aware-crowdnav/home
MIT License
168 stars 30 forks source link

Training and compute resources #1

Closed Sahil177 closed 1 year ago

Sahil177 commented 1 year ago

Hi, please could you provide information about what compute resources you used to train the model and how long it took.

Shuijing725 commented 1 year ago

We used an Intel i7-9700K CPU with 32GB memory, and an Nvidia RTX 2080 GPU. The system is any Ubuntu with version >= 16.04. The training in the default configuration takes around 48 hours. The main bottleneck is CPU speed (clock speed, number of cores) I think.

Sahil177 commented 1 year ago

Thank you very much, i noticed in arguments.py there is a option to increase the number of threads, do you recommend increasing this number to reduce this CPU bottleneck?

Shuijing725 commented 1 year ago

I tried increasing num_threads on my computer and it makes very little difference. But I'm not entirely sure how torch.set_num_threads() works. Feel free to comment here if you have any idea.