search

aharley / segaware

Segmentation-Aware Convolutional Networks Using Local Attention Masks
145 stars 31 forks source link

Check failed: error == cudaSuccess (2 vs. 0) out of memory #3

Closed zhwis closed 6 years ago

zhwis commented 6 years ago

I am currently running ./test_res.sh on a 11GB of GPU memory,gtx1080ti. But when I run the script, it immediately throws out an error: I0911 12:46:39.811416 18216 caffe.cpp:252] Running for 1449 iterations. F0911 12:46:40.284883 18216 syncedmem.cpp:56] Check failed: error == cudaSuccess (2 vs. 0) out of memory and since you are already at batch size = 1 ,I don't know how much real memory , or What should I do?

aharley commented 6 years ago

What is your image size? At 513x513 or less, it should fit in memory.

zhwis commented 6 years ago

I used voc_val.txt of VOC2012,and the maximum side is < 513,To solve this problem，I change the crop_size value in test_res.sh and segaware/config/res/test.prototxt is <=257. and I tried to resize the image smaller，but the effect is relatively poor. The problem should not be here，but I do not know how to do it

StOnEGiggity commented 6 years ago

Hi, there is the same error like @zhihuiwisdom , it looks like no detail about the environment and what type of GPU is.is It possible to share with us?Thanks a lot.

aharley commented 6 years ago

Thanks for the comments and reminders! I was using cuda-8.0, a Quadro M6000 24GB GPU, and Nvidia driver version 375.26. After the CVPR deadline (1 week) I can try to release a more memory-efficient version. I think the VGG16 version works with a 12GB gpu and no memory tricks.