search

WongKinYiu / yolov7

Implementation of paper - YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors
GNU General Public License v3.0
13.01k stars 4.11k forks source link

yolov7-e6e yolov7-w6 #1003

Open katrinacai opened 1 year ago

katrinacai commented 1 year ago

I trained YoLov7 model with the config named yolov7-e6e.yaml and size of 1280. Even though I've set the batch size to this low, I get an error "RuntimeError: CUDA out of memory." when i change yolov7-e6e.yaml to yolov7-w6.yaml with size of 1280,i still got this error. Is there something wrong with my Settings, or am I missing something?

PascAlex commented 1 year ago

Try to reduce batch size

Kannan665 commented 1 year ago

Are you using a captive GPU? And if so, what is the spec and memory? If it is 8 GB and if it is GE RTX 30 series, image size of 1280 (even with a batch size of 4) will be a tall order.......Or you must put everything in a docker container and try......Best is to use Google Colab, where you get 12 GB to 16GB GPU memory for training.......

katrinacai commented 1 year ago

Thanks for your help.I have already solved this problem.I used a wrong python file. I should run train_aux.py instead of train.py. Thanks for your help again.

------------------ 原始邮件 ------------------ 发件人: @.>; 发送时间: 2022年10月30日(星期天) 下午3:18 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [WongKinYiu/yolov7] yolov7-e6e yolov7-w6 (Issue #1003)

Are you using a captive GPU? And if so, what is the spec and memory? If it is 8 GB and if it is GE RTX 30 series, image size of 1280 (even with a batch size of 4) will be a tall order.......Or you must put everything in a docker container and try......Best is to use Google Colab, where you get 12 GB to 16GB GPU memory for training.......

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

katrinacai commented 1 year ago

Thanks for your help.It did solved my problem.Sorry,i'm too busy to replay to you. Thanks for your help again.

------------------ 原始邮件 ------------------ 发件人: @.>; 发送时间: 2022年10月27日(星期四) 下午4:15 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [WongKinYiu/yolov7] yolov7-e6e yolov7-w6 (Issue #1003)

Try to reduce batch size

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>