search

OmidPoursaeed / Generative_Adversarial_Perturbations

Generative Adversarial Perturbations (CVPR 2018)
https://arxiv.org/abs/1712.02328
135 stars 22 forks source link

Segmentation attack - GPU resources with Unet generator #6

Closed MichaKh closed 4 years ago

MichaKh commented 4 years ago

Hi,

I am trying to run your implementation for the adversarial attack on the pre-trained "fcn8s-atonce" segmentation model:

  1. According to the GAP_seg.py script arguments, the --gpu_ids flag can be set to -1 (to run it on CPU). Is the code compatible with CPU run?
  2. When trying to run the code on 2 GPUs, I am encountering "CUDA memory" issues. What is the configuration that you used for the implementation? What Pytorch/torchvision/CUDA versions, with how much CUDA memory?

Thanks

OmidPoursaeed commented 4 years ago

Hi,

I am not sure if the semantic segmentation code is compatible with CPU and can be run on multiple GPUs. We checked these with classification though, and they work.

Thanks