Jacob Logas commited on
Commit
0792228
·
unverified ·
1 Parent(s): de5d55e

Fix device usage?

Browse files
app.py CHANGED
@@ -10,7 +10,6 @@ from align.detector import detect_faces
10
  from align.align_trans import get_reference_facial_points, warp_and_crop_face
11
  import torchvision.transforms as transforms
12
 
13
- device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
14
  to_tensor = transforms.ToTensor()
15
 
16
  eps = 0.05
@@ -45,6 +44,8 @@ for root in model_roots:
45
 
46
  @spaces.GPU(duration=120)
47
  def protect(img, progress=gr.Progress(track_tqdm=True)):
 
 
48
  models_attack, V_reduction, dim = prepare_models(
49
  model_backbones,
50
  input_size,
 
10
  from align.align_trans import get_reference_facial_points, warp_and_crop_face
11
  import torchvision.transforms as transforms
12
 
 
13
  to_tensor = transforms.ToTensor()
14
 
15
  eps = 0.05
 
44
 
45
  @spaces.GPU(duration=120)
46
  def protect(img, progress=gr.Progress(track_tqdm=True)):
47
+ device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
48
+
49
  models_attack, V_reduction, dim = prepare_models(
50
  model_backbones,
51
  input_size,
util/attack_utils.py CHANGED
@@ -55,6 +55,7 @@ class Attack(nn.Module):
55
 
56
  def execute(self, images, dir_vec, direction):
57
  device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 
58
  self.lpips_loss.to(device)
59
  images = Variable(images).to(device)
60
  dir_vec = dir_vec.to(device)
 
55
 
56
  def execute(self, images, dir_vec, direction):
57
  device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
58
+ print("Device in Excute:", device)
59
  self.lpips_loss.to(device)
60
  images = Variable(images).to(device)
61
  dir_vec = dir_vec.to(device)
util/feature_extraction_utils.py CHANGED
@@ -5,11 +5,10 @@ import torchvision.transforms as transforms
5
  import torch.nn as nn
6
  import numpy as np
7
 
8
- device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
9
-
10
 
11
  def warp_image(tensor_img, theta_warp, crop_size=112):
12
  # applies affine transform theta to image and crops it
 
13
 
14
  theta_warp = torch.Tensor(theta_warp).unsqueeze(0).to(device)
15
  grid = F.affine_grid(theta_warp, tensor_img.size())
 
5
  import torch.nn as nn
6
  import numpy as np
7
 
 
 
8
 
9
  def warp_image(tensor_img, theta_warp, crop_size=112):
10
  # applies affine transform theta to image and crops it
11
+ device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
12
 
13
  theta_warp = torch.Tensor(theta_warp).unsqueeze(0).to(device)
14
  grid = F.affine_grid(theta_warp, tensor_img.size())
util/prepare_utils.py CHANGED
@@ -123,6 +123,8 @@ def get_ensemble(
123
  warp=False,
124
  theta_warp=None,
125
  ):
 
 
126
  # function prepares ensemble of feature extractors
127
  # outputs list of pytorch nn models
128
  feature_extractor_ensemble = []
@@ -213,6 +215,8 @@ def prepare_models(
213
  using_subspace,
214
  V_reduction_root,
215
  ):
 
 
216
  backbone_dict = {
217
  "IR_50": IR_50(input_size),
218
  "IR_152": IR_152(input_size),
 
123
  warp=False,
124
  theta_warp=None,
125
  ):
126
+ device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
127
+
128
  # function prepares ensemble of feature extractors
129
  # outputs list of pytorch nn models
130
  feature_extractor_ensemble = []
 
215
  using_subspace,
216
  V_reduction_root,
217
  ):
218
+ device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
219
+
220
  backbone_dict = {
221
  "IR_50": IR_50(input_size),
222
  "IR_152": IR_152(input_size),