Fix style checks

Signed-off-by: Beat Buesser <[email protected]>

Author: beat-buesser

.github/workflows/ci-lingvo.yml

@@ -50,7 +50,7 @@ jobs:
           sudo apt-get update
           sudo apt-get -y -q install ffmpeg libavcodec-extra
           python -m pip install --upgrade pip setuptools wheel
-          pip install -q -r <(sed '/^scipy/d;/^matplotlib/d;/^pandas/d;/^statsmodels/d;/^numba/d;/^jax/d;/^h5py/d;/^Pillow/d;/^pytest/d;/^pytest-mock/d;/^torch/d;/^torchaudio/d;/^torchvision/d;/^xgboost/d;/^requests/d;/^tensorflow/d;/^keras/d;/^kornia/d;/^librosa/d;/^tqdm/d;/^timm/d;/^catboost/d;/^scikit-learn/d;/^GPy/d;/^lief/d' requirements_test.txt)
+          pip install -q -r <(sed '/^scipy/d;/^matplotlib/d;/^pandas/d;/^statsmodels/d;/^numba/d;/^jax/d;/^h5py/d;/^Pillow/d;/^pytest/d;/^pytest-mock/d;/^torch/d;/^torchaudio/d;/^torchvision/d;/^xgboost/d;/^requests/d;/^tensorflow/d;/^keras/d;/^kornia/d;/^librosa/d;/^tqdm/d;/^timm/d;/^catboost/d;/^scikit-learn/d;/^GPy/d;/^lief/d;/^ultralytics/d;/^ipython/d' requirements_test.txt)
           pip install scipy==1.5.4
           pip install matplotlib==3.3.4
           pip install pandas==1.1.5

.github/workflows/ci-tensorflow-v1.yml

@@ -48,7 +48,7 @@ jobs:
           sudo apt-get update
           sudo apt-get -y -q install ffmpeg libavcodec-extra
           python -m pip install --upgrade pip setuptools wheel
-          pip install -q -r <(sed '/^pandas/d;/^scipy/d;/^matplotlib/d;/^xgboost/d;/^tensorflow/d;/^keras/d;/^jax/d;/^torch/d;/^Pillow/d;/^h5py/d;/^kornia/d;/^scikit-learn/d;/^pytest-mock/d;/^GPy/d;/^lief/d;/^statsmodels/d' requirements_test.txt)
+          pip install -q -r <(sed '/^pandas/d;/^scipy/d;/^matplotlib/d;/^xgboost/d;/^tensorflow/d;/^keras/d;/^jax/d;/^torch/d;/^Pillow/d;/^h5py/d;/^kornia/d;/^scikit-learn/d;/^pytest-mock/d;/^GPy/d;/^lief/d;/^statsmodels/d;/^ultralytics/d;/^ipython/d' requirements_test.txt)
           pip install pandas==1.3.5
           pip install scipy==1.7.2
           pip install matplotlib==3.5.3

art/attacks/evasion/overload/box_iou.py

@@ -1,3 +1,4 @@
+# pylint: disable=C0103,C0114
 #                    GNU AFFERO GENERAL PUBLIC LICENSE
 #                       Version 3, 19 November 2007
 #
@@ -660,8 +661,14 @@
 # For more information on this, and how to apply and follow the GNU AGPL, see
 # <https://www.gnu.org/licenses/>.
 
+from typing import TYPE_CHECKING
 
-def box_iou(box1: "torch.tensor", box2: "torch.tensor", eps: float = 1e-7) -> "torch.tensor":
+if TYPE_CHECKING:
+    # pylint: disable=C0412
+    import torch
+
+
+def box_iou(box1: "torch.Tensor", box2: "torch.Tensor", eps: float = 1e-7) -> "torch.Tensor":
     """
     === NOTE ===
     This function is copied from YOLOv5 repository (yolov5/utils/metrics.py)
@@ -684,4 +691,3 @@ def box_iou(box1: "torch.tensor", box2: "torch.tensor", eps: float = 1e-7) -> "t
 
     # IoU = inter / (area1 + area2 - inter)
     return inter / ((a2 - a1).prod(2) + (b2 - b1).prod(2) - inter + eps)
-

art/attacks/evasion/overload/overload.py

@@ -97,12 +97,11 @@ def generate(self, x: np.ndarray, y: Optional[np.ndarray] = None, **kwargs) -> n
 
         return x_adv
 
-    def _generate_batch(self, x_batch: np.ndarray, y_batch: Optional[np.ndarray] = None) -> np.ndarray:
+    def _generate_batch(self, x_batch: np.ndarray) -> np.ndarray:
         """
         Run the attack on a batch of images.
 
         :param x_batch: A batch of original examples.
-        :param y_batch: Not Used.
         :return: A batch of adversarial examples.
         """
 
@@ -144,7 +143,7 @@ def _attack(self, x_adv: "torch.Tensor", x: "torch.Tensor") -> "torch.Tensor":
         x_adv.requires_grad_(False)
         return x_adv
 
-    def _loss(self, x: "torch.tensor") -> Tuple["torch.tensor", "torch.tensor"]:
+    def _loss(self, x: "torch.Tensor") -> Tuple["torch.Tensor", "torch.Tensor"]:
         """
         Compute the weight of each pixel and the overload loss for a given image.
 
@@ -155,13 +154,13 @@ def _loss(self, x: "torch.tensor") -> Tuple["torch.tensor", "torch.tensor"]:
         import torch
 
         adv_logits = self.estimator.model.model(x)
-        if type(adv_logits) is tuple:
+        if isinstance(adv_logits, tuple):
             adv_logits = adv_logits[0]
 
-        THRESHOLD = self.estimator.model.conf
+        threshold = self.estimator.model.conf
         conf = adv_logits[..., 4]
         prob = adv_logits[..., 5:]
-        prob = torch.where(conf[:, :, None] * prob > THRESHOLD, torch.ones_like(prob), prob)
+        prob = torch.where(conf[:, :, None] * prob > threshold, torch.ones_like(prob), prob)
         prob = torch.sum(prob, dim=2)
         conf = conf * prob
 
@@ -174,19 +173,19 @@ def _loss(self, x: "torch.tensor") -> Tuple["torch.tensor", "torch.tensor"]:
             stride_x = x.shape[-2] // self.num_grid
             stride_y = x.shape[-1] // self.num_grid
             grid_box = torch.zeros((0, 4), device=x.device)
-            for ii in range(self.num_grid):
-                for jj in range(self.num_grid):
-                    x1 = ii * stride_x
-                    y1 = jj * stride_y
-                    x2 = min(x1 + stride_x, x.shape[-2])
-                    y2 = min(y1 + stride_y, x.shape[-1])
-                    bb = torch.as_tensor([x1, y1, x2, y2], device=x.device)[None, :]
-                    grid_box = torch.cat([grid_box, bb], dim=0)
-
-            for xi in range(x.shape[0]):
-                xyhw = adv_logits[xi, :, :4]
-                prob = torch.max(adv_logits[xi, :, 5:], dim=1).values
-                box_idx = adv_logits[xi, :, 4] * prob > THRESHOLD
+            for i_i in range(self.num_grid):
+                for j_j in range(self.num_grid):
+                    x_1 = i_i * stride_x
+                    y_1 = j_j * stride_y
+                    x_2 = min(x_1 + stride_x, x.shape[-2])
+                    y_2 = min(y_1 + stride_y, x.shape[-1])
+                    b_b = torch.as_tensor([x_1, y_1, x_2, y_2], device=x.device)[None, :]
+                    grid_box = torch.cat([grid_box, b_b], dim=0)
+
+            for x_i in range(x.shape[0]):
+                xyhw = adv_logits[x_i, :, :4]
+                prob = torch.max(adv_logits[x_i, :, 5:], dim=1).values
+                box_idx = adv_logits[x_i, :, 4] * prob > threshold
                 xyhw = xyhw[box_idx]
                 c_xyxy = self.xywh2xyxy(xyhw)
                 scores = box_iou(grid_box, c_xyxy)
@@ -197,20 +196,21 @@ def _loss(self, x: "torch.tensor") -> Tuple["torch.tensor", "torch.tensor"]:
                 # Increase the weight of the grid with fewer objects
                 idx_min = torch.argmin(scores)
                 grid_min = grid_box[idx_min]
-                x1, y1, x2, y2 = grid_min.int()
-                pixel_weight[xi, :, y1:y2, x1:x2] = pixel_weight[xi, :, y1:y2, x1:x2] * 2
-                pixel_weight = pixel_weight / torch.max(pixel_weight[xi, :]) / 255.0
+                x_1, y_1, x_2, y_2 = grid_min.int()
+                pixel_weight[x_i, :, y_1:y_2, x_1:x_2] = pixel_weight[x_i, :, y_1:y_2, x_1:x_2] * 2
+                pixel_weight = pixel_weight / torch.max(pixel_weight[x_i, :]) / 255.0
 
         return ind_loss, pixel_weight
 
-    def xywh2xyxy(self, xywh: "torch.tensor") -> "torch.tensor":
+    @staticmethod
+    def xywh2xyxy(xywh: "torch.Tensor") -> "torch.Tensor":
         """
         Convert the representation from xywh format yo xyxy format.
 
-        : param xyhw: A n by 4 boxes store the information in xyhw format
-                      where [x ,y, w h] is [center_x, center_y, width, height]
-        : return: The n by 4 boxex in xyxy format
-                  where [x1, y1, x2, y2] is [top_left_x, top_left_y, bottom_right_x,  bottom_right_y]
+        :param xyhw: A n by 4 boxes store the information in xyhw format
+                     where [x ,y, w h] is [center_x, center_y, width, height]
+        :return: The n by 4 boxes in xyxy format
+                 where [x1, y1, x2, y2] is [top_left_x, top_left_y, bottom_right_x,  bottom_right_y]
         """
         xyxy = xywh.clone()
         xyxy[:, 0] = xywh[:, 0] - xywh[:, 2] / 2

requirements_test.txt

@@ -62,6 +62,8 @@ pytest-mock~=3.14.0
 pytest-cov~=4.1.0
 requests~=2.31.0
 ultralytics==8.0.217
+ipython==8.25.0
+
 # ART
 -e .
 

tests/attacks/evasion/test_overload_attack.py

@@ -26,29 +26,25 @@
 
 logger = logging.getLogger(__name__)
 
+
 @pytest.mark.only_with_platform("pytorch")
 def test_generate(art_warning):
     try:
         import torch
-        model = torch.hub.load('ultralytics/yolov5:v7.0',  model='yolov5s')
-        py_model = PyTorchYolo(model=model,
-                               input_shape=(3, 640, 640),
-                               channels_first=True)
+
+        model = torch.hub.load("ultralytics/yolov5:v7.0", model="yolov5s")
+        py_model = PyTorchYolo(model=model, input_shape=(3, 640, 640), channels_first=True)
         # Download a sample image
         import requests
         from io import BytesIO
         from PIL import Image
 
-        TARGET = 'https://ultralytics.com/images/zidane.jpg'
-        response = requests.get(TARGET)
+        target = "https://ultralytics.com/images/zidane.jpg"
+        response = requests.get(target)
         org_img = np.asarray(Image.open(BytesIO(response.content)).resize((640, 640)))
         x = np.stack([org_img.transpose((2, 0, 1))], axis=0).astype(np.float32)
 
-        attack = OverloadPyTorch(py_model,
-                                 eps = 16.0 / 255.0,
-                                 max_iter = 10,
-                                 num_grid = 10,
-                                 batch_size = 1)
+        attack = OverloadPyTorch(py_model, eps=16.0 / 255.0, max_iter=10, num_grid=10, batch_size=1)
 
         x_adv = attack.generate(x / 255.0)
         assert x.shape == x_adv.shape
@@ -67,59 +63,26 @@ def test_generate(art_warning):
 def test_check_params(art_warning):
     try:
         import torch
-        model = torch.hub.load('ultralytics/yolov5:v7.0',  model='yolov5s')
-        py_model = PyTorchYolo(model=model,
-                               input_shape=(3, 640, 640),
-                               channels_first=True)
+
+        model = torch.hub.load("ultralytics/yolov5:v7.0", model="yolov5s")
+        py_model = PyTorchYolo(model=model, input_shape=(3, 640, 640), channels_first=True)
 
         with pytest.raises(ValueError):
-           _ = OverloadPyTorch(estimator=py_model,
-                               eps=-1.0,
-                               max_iter=5,
-                               num_grid=10,
-                               batch_size=1)
+            _ = OverloadPyTorch(estimator=py_model, eps=-1.0, max_iter=5, num_grid=10, batch_size=1)
         with pytest.raises(ValueError):
-           _ = OverloadPyTorch(estimator=py_model,
-                               eps=2.0,
-                               max_iter=5,
-                               num_grid=10,
-                               batch_size=1)
+            _ = OverloadPyTorch(estimator=py_model, eps=2.0, max_iter=5, num_grid=10, batch_size=1)
         with pytest.raises(TypeError):
-           _ = OverloadPyTorch(estimator=py_model,
-                               eps=8 / 255.0,
-                               max_iter=1.0,
-                               num_grid=10,
-                               batch_size=1)
+            _ = OverloadPyTorch(estimator=py_model, eps=8 / 255.0, max_iter=1.0, num_grid=10, batch_size=1)
         with pytest.raises(ValueError):
-           _ = OverloadPyTorch(estimator=py_model,
-                               eps=8 / 255.0,
-                               max_iter=0,
-                               num_grid=10,
-                               batch_size=1)
+            _ = OverloadPyTorch(estimator=py_model, eps=8 / 255.0, max_iter=0, num_grid=10, batch_size=1)
         with pytest.raises(TypeError):
-           _ = OverloadPyTorch(estimator=py_model,
-                               eps=8 / 255.0,
-                               max_iter=5,
-                               num_grid=1.0,
-                               batch_size=1)
+            _ = OverloadPyTorch(estimator=py_model, eps=8 / 255.0, max_iter=5, num_grid=1.0, batch_size=1)
         with pytest.raises(ValueError):
-           _ = OverloadPyTorch(estimator=py_model,
-                               eps=8 / 255.0,
-                               max_iter=5,
-                               num_grid=0,
-                               batch_size=1)
+            _ = OverloadPyTorch(estimator=py_model, eps=8 / 255.0, max_iter=5, num_grid=0, batch_size=1)
         with pytest.raises(TypeError):
-           _ = OverloadPyTorch(estimator=py_model,
-                               eps=8 / 255.0,
-                               max_iter=5,
-                               num_grid=10,
-                               batch_size=1.0)
+            _ = OverloadPyTorch(estimator=py_model, eps=8 / 255.0, max_iter=5, num_grid=10, batch_size=1.0)
         with pytest.raises(ValueError):
-           _ = OverloadPyTorch(estimator=py_model,
-                               eps=8 / 255.0,
-                               max_iter=5,
-                               num_grid=0,
-                               batch_size=0)
+            _ = OverloadPyTorch(estimator=py_model, eps=8 / 255.0, max_iter=5, num_grid=0, batch_size=0)
 
     except ARTTestException as e:
         art_warning(e)