How can we use only 1 GPU for training and testing? For the current code implementation, needed 2 GPUs.

[HongminWu]

[CZ-Wu]

Sorry for my late reply. The current version only supports 2 GPUs for acceleration in grasp proposal phase. I think it's ok to run on one GPU by setting larger forward number at this point:
https://github.com/CZ-Wu/GPNet/blob/59419f4505ca0732316a7d993bfe065547892847/tools/proposal.py#L384
and setting smaller batch size at these two points:
https://github.com/CZ-Wu/GPNet/blob/59419f4505ca0732316a7d993bfe065547892847/lib/gpnet.py#L345
https://github.com/CZ-Wu/GPNet/blob/59419f4505ca0732316a7d993bfe065547892847/lib/gpnet.py#L404
eg. change 5000 to 2000 or smaller.

[HongminWu]

Got it. Thanks for your feedback. ----------------------------------------- Hongmin Wu, Post-doc Project Leader of Robots Team Guangdong Institute of Intelligent Manufacturing (GIIM) Guangzhou, Guangdong, P.R.China 510006 WeChat: Homing20 Facebook: Hongmin Wu Tel: 86+18819498204 E-mail: *hongminwu0120@gmail.com <hongminwu0120@gmail.com>, hm.wu@giim.ac.cn <hm.wu@giim.ac.cn>* Open Source: https://github.com/HongminWu

…

On Sat, 31 Oct 2020 at 15:36, CZ-Wu ***@***.***> wrote: Sorry for my late reply. The current version only supports 2 GPUs for acceleration in grasp proposal phase. I think it's ok to run on one GPU by setting larger forward number at this point: https://github.com/CZ-Wu/GPNet/blob/59419f4505ca0732316a7d993bfe065547892847/tools/proposal.py#L384 and setting smaller batch size at these two points: https://github.com/CZ-Wu/GPNet/blob/59419f4505ca0732316a7d993bfe065547892847/lib/gpnet.py#L345 https://github.com/CZ-Wu/GPNet/blob/59419f4505ca0732316a7d993bfe065547892847/lib/gpnet.py#L404 eg. change 5000 to 2000 or smaller. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <https://github.com/CZ-Wu/GPNet/issues/1#issuecomment-719897995>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACZNS6KXVP7EWSO6UUAZDJLSNO5ARANCNFSM4S6CG4XA> .

[HongminWu]

font{ line-height: 1.6; } ul,ol{ padding-left: 20px; list-style-position: inside; } One more question, when do you update the simulation environment on Github using pybullet for more intuitive comprehension of your work?  Best regards,-- Hongmin Wu, Post-docProject Leader of Robots TeamGuangdong Institute of Intelligent Manufacturing (GIIM)Guangzhou, Guangdong, P.R.China 510006WeChat: Homing20Facebook: Hongmin WuTel: 86+18819498204E-mail: hongminwu0120@gmail.com, hm.wu@giim.ac.cnOpen Source: https://github.com/HongminWu On 11/2/2020 09:04，Hongmin Wu<hongminwu0120@gmail.com> wrote： Got it. Thanks for your feedback.-----------------------------------------Hongmin Wu, Post-docProject Leader of Robots TeamGuangdong Institute of Intelligent Manufacturing (GIIM)Guangzhou, Guangdong, P.R.China 510006WeChat: Homing20Facebook: Hongmin WuTel: 86+18819498204E-mail: hongminwu0120@gmail.com, hm.wu@giim.ac.cnOpen Source: https://github.com/HongminWuOn Sat, 31 Oct 2020 at 15:36, CZ-Wu <notifications@github.com> wrote: Sorry for my late reply. The current version only supports 2 GPUs for acceleration in grasp proposal phase. I think it's ok to run on one GPU by setting larger forward number at this point: https://github.com/CZ-Wu/GPNet/blob/59419f4505ca0732316a7d993bfe065547892847/tools/proposal.py#L384 and setting smaller batch size at these two points: https://github.com/CZ-Wu/GPNet/blob/59419f4505ca0732316a7d993bfe065547892847/lib/gpnet.py#L345 https://github.com/CZ-Wu/GPNet/blob/59419f4505ca0732316a7d993bfe065547892847/lib/gpnet.py#L404 eg. change 5000 to 2000 or smaller. —You are receiving this because you authored the thread.Reply to this email directly, view it on GitHub, or unsubscribe.