search

NVIDIA / FastPhotoStyle

Style transfer, deep learning, feature transform
Other
11.11k stars 1.2k forks source link

2080ti run demopy :cuda runtime error(11) #71

Open xujin1184104394 opened 5 years ago

xujin1184104394 commented 5 years ago

environment: CUDA10 CUDNN 7.4.2 Pytorch 1.0.1 GPU 2080ti Run python demo.py to report an error:

python demo.py --output_image_path results/example1.png Resize image: (803,538)->(803,538) Resize image: (960,540)->(960,540) THCudaCheck FAIL file=/pytorch/aten/src/THC/THCGeneral.cpp line=405 error=11 : invalid argument Elapsed time in stylization: 0.300527 Traceback (most recent call last): File "demo.py", line 47, in no_post=args.no_post File "/home1/code/FastPhotoStyle/process_stylization.py", line 122, in stylization stylized_img = stylization_module.transform(cont_img, styl_img, cont_seg, styl_seg) File "/home1/code/FastPhotoStyle/photo_wct.py", line 52, in transform csF1 = self.feature_wct(cF1, sF1, cont_seg, styl_seg) File "/home1/code/FastPhotoStyle/photo_wct.py", line 77, in feature_wct target_feature = self.wct_core(cont_feat_view, styl_feat_view) File "/home1/code/FastPhotoStyle/photo_wct.py", line 140, in wct_core if c_e[i] >= 0.00001: RuntimeError: cuda runtime error (11) : invalid argument at /pytorch/aten/src/THC/generated/../THCTensorMathCompare.cuh:82

leijuzi commented 5 years ago

i meet the same issue are you solve it now?

xujin1184104394 commented 5 years ago

No

jbohnslav commented 5 years ago

I had a similar issue on my new RTX GPU. Running on the 1080Ti was fine, and I got a bug in THC with the RTX. I installed PyTorch from source (Ubuntu 18.04), which solved my issue.

zrongcheng commented 4 years ago

Please use the cuda 10.0.