search

bobbens / sketch_simplification

Models and code related to sketch simplification of rough sketches.
https://esslab.jp/~ess/research/sketch_master/
Other
726 stars 103 forks source link

GPU size error? #1

Open ygren opened 6 years ago

ygren commented 6 years ago

when I execute this command: python simplify.py the error as follows: THCudaCheck FAIL file=/pytorch/torch/lib/THC/generic/THCStorage.cu line=58 error=2 : out of memory

My GPU is 1080TI 11G, can you tell me how much computing resources are required for your model? Thanks!

bobbens commented 6 years ago

Dependent on input image resolution size, but with 11G you should be able to handle fairly large images of at least 1024x1024 although likely quite a bit more. Make sure you have latest pytorch/cuda/cudnn (cudnn makes a big difference in my experience).

b4zz4 commented 6 years ago

The same problem with nvidia K1000

Processing fig06_danshi.png...THCudaCheck FAIL file=/pytorch/torch/lib/THC/generic/THCStorage.c line=82 error=2 : out of memory ./figs.sh: línea 3: 22867 Violación de segmento python simplify.py --img "$IN" --out "$OUT"

image size is 170x224

ygren commented 6 years ago

@bobbens Thanks! I have run it on my MACBOOK PRO without CUDA，and the python version is 3.6.maybe the reason is python version is 3.6 not 2.7.It's a very amazing job,but the effect is still not ideal. Do you have a plan to release your code of your network structure? I want to do some work on your basis.

DifanLiu commented 6 years ago

The same problem with nvidia K5200 THCudaCheck FAIL file=/opt/conda/conda-bld/pytorch_1503966894950/work/torch/lib/THC/generic/THCStorage.c line=79 error=2 : out of memory What's your pytorch/cuda/cudnn version? @bobbens Could you please release your whole code of your project, including the discriminator? Thank you so much!

bobbens commented 6 years ago

@YgRen the network structure is actually included in the .t7 file. Most of my tests have been done using Cuda 9.1, Cudnn 7.5, and a GTX 1080. Using older cards will probably hit memory issues as apparently the torch7 code is much more memory intensive. When I get some time I'll try converting to native pytorch which should be more efficient.

laoxihongshi commented 6 years ago

i have using Cuda 9.1, Cudnn 7.5, and a GTX 1080，but when i run all the demo cmd， i have the problem “out of memory” plz help me

Gabrielnero000 commented 5 years ago

GTX 1050 Ti gives the same error (RuntimeError: CUDA error: out of memory), someone solve it?