Merge pull request #2192 from kieranfraser/dev_detection_transformer

Detection Transformer Estimator

.github/workflows/ci-pytorch-object-detectors.yml

@@ -50,6 +50,8 @@ jobs:
         run: pytest --cov-report=xml --cov=art --cov-append -q -vv tests/estimators/object_detection/test_pytorch_object_detector.py --framework=pytorch --durations=0
       - name: Run Test Action - test_pytorch_faster_rcnn
         run: pytest --cov-report=xml --cov=art --cov-append -q -vv tests/estimators/object_detection/test_pytorch_faster_rcnn.py --framework=pytorch --durations=0
+      - name: Run Test Action - test_pytorch_detection_transformer
+        run: pytest --cov-report=xml --cov=art --cov-append -q -vv tests/estimators/object_detection/test_pytorch_detection_transformer.py --framework=pytorch --durations=0
       - name: Upload coverage to Codecov
         uses: codecov/codecov-action@v3
         with:

art/estimators/object_detection/__init__.py

@@ -8,3 +8,4 @@
 from art.estimators.object_detection.pytorch_yolo import PyTorchYolo
 from art.estimators.object_detection.tensorflow_faster_rcnn import TensorFlowFasterRCNN
 from art.estimators.object_detection.tensorflow_v2_faster_rcnn import TensorFlowV2FasterRCNN
+from art.estimators.object_detection.pytorch_detection_transformer import PyTorchDetectionTransformer

art/estimators/pytorch.py

@@ -329,3 +329,14 @@ def set_batchnorm(self, train: bool) -> None:
 
         # pylint: disable=W0212
         self._set_layer(train=train, layerinfo=[torch.nn.modules.batchnorm._BatchNorm])  # type: ignore
+
+    def set_multihead_attention(self, train: bool) -> None:
+        """
+        Set all multi-head attention layers into train or eval mode.
+
+        :param train: False for evaluation mode.
+        """
+        import torch
+
+        # pylint: disable=W0212
+        self._set_layer(train=train, layerinfo=[torch.nn.modules.MultiheadAttention])  # type: ignore

notebooks/README.md

@@ -66,6 +66,10 @@ models.
 [attack_adversarial_patch_faster_rcnn.ipynb](adversarial_patch/attack_adversarial_patch_faster_rcnn.ipynb) [[on nbviewer](https://nbviewer.jupyter.org/github/Trusted-AI/adversarial-robustness-toolbox/blob/main/notebooks/adversarial_patch/attack_adversarial_patch_faster_rcnn.ipynb)]
 shows how to set up a TFv2 Faster R-CNN object detector with ART and create an adversarial patch attack that fools the detector.
 
+[attack_adversarial_patch_detr.ipynb](adversarial_patch/attack_adversarial_patch_detr.ipynb) [[on nbviewer](https://nbviewer.jupyter.org/github/Trusted-AI/adversarial-robustness-toolbox/blob/main/notebooks/adversarial_patch/attack_adversarial_patch_detr.ipynb)]
+shows how to set up the DEtection TRansformer (DETR) with ART for object detection and create an adversarial patch attack that fools the detector.
+
+
 <p align="center">
   <img src="../utils/data/images/adversarial_patch.png?raw=true" width="200" title="adversarial_patch">
 </p>

tests/estimators/object_detection/test_pytorch_detection_transformer.py

@@ -0,0 +1,309 @@
+# MIT License
+#
+# Copyright (C) The Adversarial Robustness Toolbox (ART) Authors 2023
+#
+# Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated
+# documentation files (the "Software"), to deal in the Software without restriction, including without limitation the
+# rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit
+# persons to whom the Software is furnished to do so, subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in all copies or substantial portions of the
+# Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE
+# WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+# TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+# SOFTWARE.
+from __future__ import absolute_import, division, print_function, unicode_literals
+
+import logging
+
+import numpy as np
+import pytest
+
+logger = logging.getLogger(__name__)
+
+
+@pytest.fixture()
+@pytest.mark.skip_framework("tensorflow", "tensorflow2v1", "keras", "kerastf", "mxnet", "non_dl_frameworks")
+def get_pytorch_detr():
+    from art.utils import load_dataset
+    from art.estimators.object_detection.pytorch_detection_transformer import PyTorchDetectionTransformer
+
+    MEAN = [0.485, 0.456, 0.406]
+    STD = [0.229, 0.224, 0.225]
+    INPUT_SHAPE = (3, 32, 32)
+
+    object_detector = PyTorchDetectionTransformer(
+        input_shape=INPUT_SHAPE, clip_values=(0, 1), preprocessing=(MEAN, STD)
+    )
+
+    n_test = 2
+    (_, _), (x_test, y_test), _, _ = load_dataset("cifar10")
+    x_test = x_test.transpose(0, 3, 1, 2).astype(np.float32)
+    x_test = x_test[:n_test]
+
+    # Create labels
+
+    result = object_detector.predict(x=x_test)
+
+    y_test = [
+        {
+            "boxes": result[0]["boxes"],
+            "labels": result[0]["labels"],
+            "scores": np.ones_like(result[0]["labels"]),
+        },
+        {
+            "boxes": result[1]["boxes"],
+            "labels": result[1]["labels"],
+            "scores": np.ones_like(result[1]["labels"]),
+        },
+    ]
+
+    yield object_detector, x_test, y_test
+
+
+@pytest.mark.only_with_platform("pytorch")
+def test_predict(get_pytorch_detr):
+
+    object_detector, x_test, _ = get_pytorch_detr
+
+    result = object_detector.predict(x=x_test)
+
+    assert list(result[0].keys()) == ["boxes", "labels", "scores"]
+
+    assert result[0]["boxes"].shape == (100, 4)
+    expected_detection_boxes = np.asarray([-5.9490204e-03, 1.1947733e01, 3.1993944e01, 3.1925127e01])
+    np.testing.assert_array_almost_equal(result[0]["boxes"][2, :], expected_detection_boxes, decimal=1)
+
+    assert result[0]["scores"].shape == (100,)
+    expected_detection_scores = np.asarray(
+        [
+            0.00679839,
+            0.0250559,
+            0.07205943,
+            0.01115368,
+            0.03321039,
+            0.10407761,
+            0.00113309,
+            0.01442852,
+            0.00527624,
+            0.01240906,
+        ]
+    )
+    np.testing.assert_array_almost_equal(result[0]["scores"][:10], expected_detection_scores, decimal=1)
+
+    assert result[0]["labels"].shape == (100,)
+    expected_detection_classes = np.asarray([17, 17, 33, 17, 17, 17, 74, 17, 17, 17])
+    np.testing.assert_array_almost_equal(result[0]["labels"][:10], expected_detection_classes, decimal=5)
+
+
+@pytest.mark.only_with_platform("pytorch")
+def test_loss_gradient(get_pytorch_detr):
+
+    object_detector, x_test, y_test = get_pytorch_detr
+
+    grads = object_detector.loss_gradient(x=x_test, y=y_test)
+
+    assert grads.shape == (2, 3, 800, 800)
+
+    expected_gradients1 = np.asarray(
+        [
+            -0.00061366,
+            0.00322502,
+            -0.00039866,
+            -0.00807413,
+            -0.00476555,
+            0.00181204,
+            0.01007765,
+            0.00415828,
+            -0.00073114,
+            0.00018387,
+            -0.00146992,
+            -0.00119636,
+            -0.00098966,
+            -0.00295517,
+            -0.0024271,
+            -0.00131314,
+            -0.00149217,
+            -0.00104926,
+            -0.00154239,
+            -0.00110989,
+            0.00092887,
+            0.00049146,
+            -0.00292508,
+            -0.00124526,
+            0.00140347,
+            0.00019833,
+            0.00191074,
+            -0.00117537,
+            -0.00080604,
+            0.00057427,
+            -0.00061728,
+            -0.00206535,
+        ]
+    )
+
+    np.testing.assert_array_almost_equal(grads[0, 0, 10, :32], expected_gradients1, decimal=2)
+
+    expected_gradients2 = np.asarray(
+        [
+            -1.1787530e-03,
+            -2.8500680e-03,
+            5.0884970e-03,
+            6.4504531e-04,
+            -6.8841036e-05,
+            2.8184296e-03,
+            3.0257765e-03,
+            2.8565727e-04,
+            -1.0701057e-04,
+            1.2945699e-03,
+            7.3593057e-04,
+            1.0177144e-03,
+            -2.4692707e-03,
+            -1.3801848e-03,
+            6.3182280e-04,
+            -4.2305476e-04,
+            4.4307750e-04,
+            8.5821096e-04,
+            -7.1204413e-04,
+            -3.1404425e-03,
+            -1.5964351e-03,
+            -1.9222996e-03,
+            -5.3157361e-04,
+            -9.9202688e-04,
+            -1.5815455e-03,
+            2.0060266e-04,
+            -2.0584739e-03,
+            6.6960667e-04,
+            9.7393827e-04,
+            -1.6040013e-03,
+            -6.9741381e-04,
+            1.4657658e-04,
+        ]
+    )
+    np.testing.assert_array_almost_equal(grads[1, 0, 10, :32], expected_gradients2, decimal=2)
+
+
+@pytest.mark.only_with_platform("pytorch")
+def test_errors():
+
+    from torch import hub
+
+    from art.estimators.object_detection.pytorch_detection_transformer import PyTorchDetectionTransformer
+
+    model = hub.load("facebookresearch/detr", "detr_resnet50", pretrained=True)
+
+    with pytest.raises(ValueError):
+        PyTorchDetectionTransformer(
+            model=model,
+            clip_values=(1, 2),
+            attack_losses=("loss_ce", "loss_bbox", "loss_giou"),
+        )
+
+    with pytest.raises(ValueError):
+        PyTorchDetectionTransformer(
+            model=model,
+            clip_values=(-1, 1),
+            attack_losses=("loss_ce", "loss_bbox", "loss_giou"),
+        )
+
+    from art.defences.postprocessor.rounded import Rounded
+
+    post_def = Rounded()
+    with pytest.raises(ValueError):
+        PyTorchDetectionTransformer(
+            model=model,
+            clip_values=(0, 1),
+            attack_losses=("loss_ce", "loss_bbox", "loss_giou"),
+            postprocessing_defences=post_def,
+        )
+
+
+@pytest.mark.only_with_platform("pytorch")
+def test_preprocessing_defences(get_pytorch_detr):
+
+    object_detector, x_test, _ = get_pytorch_detr
+
+    from art.defences.preprocessor.spatial_smoothing_pytorch import SpatialSmoothingPyTorch
+
+    pre_def = SpatialSmoothingPyTorch()
+
+    object_detector.set_params(preprocessing_defences=pre_def)
+
+    # Create labels
+    result = object_detector.predict(x=x_test)
+
+    y = [
+        {
+            "boxes": result[0]["boxes"],
+            "labels": result[0]["labels"],
+            "scores": np.ones_like(result[0]["labels"]),
+        },
+        {
+            "boxes": result[1]["boxes"],
+            "labels": result[1]["labels"],
+            "scores": np.ones_like(result[1]["labels"]),
+        },
+    ]
+
+    # Compute gradients
+    grads = object_detector.loss_gradient(x=x_test, y=y)
+
+    assert grads.shape == (2, 3, 800, 800)
+
+
+@pytest.mark.only_with_platform("pytorch")
+def test_compute_losses(get_pytorch_detr):
+
+    object_detector, x_test, y_test = get_pytorch_detr
+    object_detector.attack_losses = "loss_ce"
+    losses = object_detector.compute_losses(x=x_test, y=y_test)
+    assert len(losses) == 1
+
+
+@pytest.mark.only_with_platform("pytorch")
+def test_compute_loss(get_pytorch_detr):
+
+    object_detector, x_test, _ = get_pytorch_detr
+    # Create labels
+    result = object_detector.predict(x_test)
+
+    y = [
+        {
+            "boxes": result[0]["boxes"],
+            "labels": result[0]["labels"],
+            "scores": np.ones_like(result[0]["labels"]),
+        },
+        {
+            "boxes": result[1]["boxes"],
+            "labels": result[1]["labels"],
+            "scores": np.ones_like(result[1]["labels"]),
+        },
+    ]
+
+    # Compute loss
+    loss = object_detector.compute_loss(x=x_test, y=y)
+
+    assert pytest.approx(3.9634, abs=0.01) == float(loss)
+
+
+@pytest.mark.only_with_platform("pytorch")
+def test_pgd(get_pytorch_detr):
+
+    object_detector, x_test, y_test = get_pytorch_detr
+
+    from art.attacks.evasion import ProjectedGradientDescent
+    from PIL import Image
+
+    imgs = []
+    for i in x_test:
+        img = Image.fromarray((i * 255).astype(np.uint8).transpose(1, 2, 0))
+        img = img.resize(size=(800, 800))
+        imgs.append(np.array(img))
+    x_test = np.array(imgs).transpose(0, 3, 1, 2)
+
+    attack = ProjectedGradientDescent(estimator=object_detector, max_iter=2)
+    x_test_adv = attack.generate(x=x_test, y=y_test)
+    np.testing.assert_raises(AssertionError, np.testing.assert_array_equal, x_test_adv, x_test)