how long should I wait for the first train step (the Farthest Point Sampling model) to be done ?

[vancewangdj]

Dear Sir

 Nice to meet you, my name is vance. 
 I already setup the python development environment(windows) and finished preprocess.

 I am running the train-1 (the Farthest Point Sampling model) around 16 hours with my notebook and still waiting for the finished of the first train step. 

 I am just wondering how long should I wait for the first train step to be done ?

[limhoyeon]

It took 1 hour for training one step(1200 mesh object) with 2080rtx in my case. Thanks

2024년 3월 1일 (금) 오전 10:37, vancewangdj @.***>님이 작성:

Dear Sir

Nice to meet you, my name is vance. I already setup the python development environment(windows) and finished preprocess.

I am running the train-1 (the Farthest Point Sampling model) around 16 hours with my notebook and still waiting for the finished of the first train step.

I am just wondering how long should I wait for the first train step to be done ?

messageImage_1709256984210.jpg (view on web) https://github.com/limhoyeon/ToothGroupNetwork/assets/120362498/bb1e9c4d-0c6b-4d5f-9153-70c416f8e24b

— Reply to this email directly, view it on GitHub https://github.com/limhoyeon/ToothGroupNetwork/issues/27, or unsubscribe https://github.com/notifications/unsubscribe-auth/AQW6T2QUX7V5DO7UAEMMCIDYV7LVNAVCNFSM6AAAAABEA6NZCSVHI2DSMVQWIX3LMV43ASLTON2WKOZSGE3DEMZZGUZTENY . You are receiving this because you are subscribed to this thread.Message ID: @.***>

[vancewangdj]

so do you mean i can stop the train one step(1200 mesh object) after one hour ??

i don't need to run the train one step too long, right ?

so i can stop the train one step after 1 hour, if the train one result is already convergent ?

[vancewangdj]

by the way my graphic card is rtx 2070 with max-q design

[vancewangdj]

so you just run 1 hour for training the Farthest Point Sampling model ?

[aanandaathmaram]

Hi Sir @vancewangdj , Could you guide me in setting up this project? I have been facing errors related to pointops_cuda and others, even though I have replaced them with the ones provided by the author @limhoyeon .

[limhoyeon]

I think that replacing point ops library in "https://github.com/POSTECH-CVLab/point-transformer" can help you

Thank you

[limhoyeon]

@vancewangdj yes, It takes one hour for one step(training 1200 models). And I trained 60 steps.