RenYurui / Global-Flow-Local-Attention

The source code for paper "Deep Image Spatial Transformation for Person Image Generation"
https://renyurui.github.io/GFLA-web
Other
568 stars 84 forks source link

Domo Pose-Guided Person Image Generation not working #58

Closed Peu-m closed 4 years ago

Peu-m commented 4 years ago

Hello,

I'm unable to run demo Person Image Generation.

here is my colab GPU architecture:

+-----------------------------------------------------------------------------+ | NVIDIA-SMI 418.67 Driver Version: 418.67 CUDA Version: 10.1 | |-------------------------------+----------------------+----------------------+ | GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. | |===============================+======================+======================| | 0 Tesla T4 Off | 00000000:00:04.0 Off | 0 | | N/A 36C P8 10W / 70W | 0MiB / 15079MiB | 0% Default | +-------------------------------+----------------------+----------------------+

I'm getting following error:

Loading data pairs ... Loading data pairs finished ... dataset [FashionDataset] of size 41 was created val images = 41 Network [PoseGenerator] was created. Total number of parameters: 14.0474 million. To see the architecture, do print(network). THCudaCheck FAIL file=/pytorch/aten/src/THC/generic/THCTensorMathPairwise.cu line=225 error=11 : invalid argument Traceback (most recent call last): File "demo.py", line 15, in model = create_model(opt)

RuntimeError: cuda runtime error (11) : invalid argument at /pytorch/aten/src/THC/generic/THCTensorMathPairwise.cu:225

image

Could you please suggest on how I can resolve this ?

RenYurui commented 4 years ago

The code has been tested with cuda9 and cuda10.0. I am not sure whether pytorch 1.0.0 can run with cuda 10.1.

Peu-m commented 4 years ago

Thank! for your prompt reply.

However, i also checked by installing cuda 10.0, issue remain same:

RuntimeError: cuda runtime error (11) : invalid argument at /pytorch/aten/src/THC/generic/THCTensorMathPairwise.cu:225

Original demo also uses Cuda 10.1 (screenshot attached) image

Therefore is the issue with GPU T-4 ?

Note: The current code is tested with Tesla V100. If you use a different GPU, you may need to select correct nvcc_args for your GPU when you buil Custom CUDA Extensions. Comment or Uncomment --gencode in block_extractor/setup.py, local_attn_reshape/setup.py, and resample2d_package/setup.py. Please check here for details.

RenYurui commented 4 years ago

Hi, I think it is a pytorch/cuda issue. Please check your env (cuda10 and pytorch 1.0.0 for cuda10). If the problem persists see if you can run basic straight PyTorch examples such as https://pytorch.org/tutorials/beginner/blitz/cifar10_tutorial.html. You may also need to set torch.backends.cudnn.benchmark to False if you do not use cudnn.

Peu-m commented 4 years ago

Thank You! for your reply, i will soon try and let you know on the above issue.

Peu-m commented 4 years ago

@RenYurui i tried to reset my colab env to get Tesla P100 GPU and by luck i got this GPU allocated for my runtime and found that code is working without any Error. Therefore, i think it is not pytorch and Cuda issue rather GPU architecture related issue.

I'm closing this issue as the demo code is working for Tesla P100 GPU.

Thank You