search

yhw-yhw / D2HC-RMVSNet

The official repository of the paper "Dense Hybrid Recurrent Multi-view Stereo Net with Dynamic Consistency Checking" (ECCV2020 Spotlight)
MIT License
109 stars 9 forks source link

CUDA out of memory #8

Open leno0903 opened 3 years ago

leno0903 commented 3 years ago

1080Ti * 4， batch_size = 4 RuntimeError: CUDA out of memory. Tried to allocate 20.00 MiB (GPU 0; 10.92 GiB total capacity; 10.09 GiB already allocated; 5.00 MiB free; 19.09 MiB cached)

weizizhuang commented 3 years ago

At least 18G memory is required for training phase, our models are trained on 24G Titan X graphical cards. If you only have 12G 1080Ti graphical cards, I suggest you to decrease the settings for training, e.g. number of plane sweep, number of neighbor views in train.sh.

leno0903 commented 3 years ago

thank you

agenthong commented 3 years ago

At least 18G memory is required for training phase, our models are trained on 24G Titan X graphical cards. If you only have 12G 1080Ti graphical cards, I suggest you to decrease the settings for training, e.g. number of plane sweep, number of neighbor views in train.sh.

But it's said that there is 6.6G consumed with 1200*1600 images input in the paper.

leno0903 commented 3 years ago

It's test stage

------------------ 原始邮件 ------------------ 发件人: @.>; 发送时间: 2021年5月18日(星期二) 晚上6:49 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [yhw-yhw/D2HC-RMVSNet] CUDA out of memory (#8)

At least 18G memory is required for training phase, our models are trained on 24G Titan X graphical cards. If you only have 12G 1080Ti graphical cards, I suggest you to decrease the settings for training, e.g. number of plane sweep, number of neighbor views in train.sh.

But it's said that there is 6.6G consumed with 1200*1600 images input in the paper.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.

agenthong commented 3 years ago

It's test stage … ------------------ 原始邮件 ------------------ 发件人: @.>; 发送时间: 2021年5月18日(星期二) 晚上6:49 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [yhw-yhw/D2HC-RMVSNet] CUDA out of memory (#8) At least 18G memory is required for training phase, our models are trained on 24G Titan X graphical cards. If you only have 12G 1080Ti graphical cards, I suggest you to decrease the settings for training, e.g. number of plane sweep, number of neighbor views in train.sh. But it's said that there is 6.6G consumed with 1200*1600 images input in the paper. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.

Thanks, how much memory consumed if batchsize is 1