!2314 Implemented RandomHorizontalFlipWithBBox and BoundingBoxAugment C++ Ops

Merge pull request !2314 from imaaamin/object_ops_pr

mindspore/ccsrc/dataset/api/python_bindings.cc

@@ -31,6 +31,7 @@
 #include "dataset/kernels/image/random_crop_and_resize_op.h"
 #include "dataset/kernels/image/random_crop_op.h"
 #include "dataset/kernels/image/random_horizontal_flip_op.h"
+#include "dataset/kernels/image/random_horizontal_flip_bbox_op.h"
 #include "dataset/kernels/image/random_resize_op.h"
 #include "dataset/kernels/image/random_rotation_op.h"
 #include "dataset/kernels/image/random_vertical_flip_op.h"
@@ -38,6 +39,7 @@
 #include "dataset/kernels/image/resize_bilinear_op.h"
 #include "dataset/kernels/image/resize_op.h"
 #include "dataset/kernels/image/uniform_aug_op.h"
+#include "dataset/kernels/image/bounding_box_augment_op.h"
 #include "dataset/kernels/data/fill_op.h"
 #include "dataset/kernels/data/mask_op.h"
 #include "dataset/kernels/data/pad_end_op.h"
@@ -347,6 +349,11 @@ void bindTensorOps1(py::module *m) {
     .def(py::init<std::vector<std::shared_ptr<TensorOp>>, int32_t>(), py::arg("operations"),
          py::arg("NumOps") = UniformAugOp::kDefNumOps);
 
+  (void)py::class_<BoundingBoxAugOp, TensorOp, std::shared_ptr<BoundingBoxAugOp>>(
+    *m, "BoundingBoxAugOp", "Tensor operation to apply a transformation on a random choice of bounding boxes.")
+    .def(py::init<std::shared_ptr<TensorOp>, float>(), py::arg("transform"),
+         py::arg("ratio") = BoundingBoxAugOp::defRatio);
+
   (void)py::class_<ResizeBilinearOp, TensorOp, std::shared_ptr<ResizeBilinearOp>>(
     *m, "ResizeBilinearOp",
     "Tensor operation to resize an image using "
@@ -361,6 +368,11 @@ void bindTensorOps1(py::module *m) {
   (void)py::class_<RandomHorizontalFlipOp, TensorOp, std::shared_ptr<RandomHorizontalFlipOp>>(
     *m, "RandomHorizontalFlipOp", "Tensor operation to randomly flip an image horizontally.")
     .def(py::init<float>(), py::arg("probability") = RandomHorizontalFlipOp::kDefProbability);
+
+  (void)py::class_<RandomHorizontalFlipWithBBoxOp, TensorOp, std::shared_ptr<RandomHorizontalFlipWithBBoxOp>>(
+    *m, "RandomHorizontalFlipWithBBoxOp",
+    "Tensor operation to randomly flip an image horizontally, while flipping bounding boxes.")
+    .def(py::init<float>(), py::arg("probability") = RandomHorizontalFlipWithBBoxOp::kDefProbability);
 }
 
 void bindTensorOps2(py::module *m) {

mindspore/ccsrc/dataset/kernels/image/CMakeLists.txt

@@ -13,6 +13,8 @@ add_library(kernels-image OBJECT
     random_crop_and_resize_op.cc
     random_crop_op.cc
     random_horizontal_flip_op.cc
+        random_horizontal_flip_bbox_op.cc
+        bounding_box_augment_op.cc
     random_resize_op.cc
     random_rotation_op.cc
     random_vertical_flip_op.cc

mindspore/ccsrc/dataset/kernels/image/bounding_box_augment_op.cc

@@ -0,0 +1,77 @@
+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <vector>
+#include <utility>
+#include "dataset/kernels/image/bounding_box_augment_op.h"
+#include "dataset/kernels/image/resize_op.h"
+#include "dataset/kernels/image/image_utils.h"
+#include "dataset/core/cv_tensor.h"
+
+namespace mindspore {
+namespace dataset {
+const float BoundingBoxAugOp::defRatio = 0.3;
+
+BoundingBoxAugOp::BoundingBoxAugOp(std::shared_ptr<TensorOp> transform, float ratio)
+    : ratio_(ratio), transform_(std::move(transform)) {}
+
+Status BoundingBoxAugOp::Compute(const TensorRow &input, TensorRow *output) {
+  IO_CHECK_VECTOR(input, output);
+  BOUNDING_BOX_CHECK(input);  // check if bounding boxes are valid
+  uint32_t num_of_boxes = input[1]->shape()[0];
+  uint32_t num_to_aug = num_of_boxes * ratio_;  // cast to int
+  std::vector<uint32_t> boxes(num_of_boxes);
+  std::vector<uint32_t> selected_boxes;
+  for (uint32_t i = 0; i < num_of_boxes; i++) boxes[i] = i;
+  // sample bboxes according to ratio picked by user
+  std::random_device rd;
+  std::sample(boxes.begin(), boxes.end(), std::back_inserter(selected_boxes), num_to_aug, std::mt19937(rd()));
+  std::shared_ptr<Tensor> crop_out;
+  std::shared_ptr<Tensor> res_out;
+  std::shared_ptr<CVTensor> input_restore = CVTensor::AsCVTensor(input[0]);
+
+  for (uint32_t i = 0; i < num_to_aug; i++) {
+    uint32_t min_x = 0;
+    uint32_t min_y = 0;
+    uint32_t b_w = 0;
+    uint32_t b_h = 0;
+    // get the required items
+    input[1]->GetItemAt<uint32_t>(&min_x, {selected_boxes[i], 0});
+    input[1]->GetItemAt<uint32_t>(&min_y, {selected_boxes[i], 1});
+    input[1]->GetItemAt<uint32_t>(&b_w, {selected_boxes[i], 2});
+    input[1]->GetItemAt<uint32_t>(&b_h, {selected_boxes[i], 3});
+    Crop(input_restore, &crop_out, min_x, min_y, b_w, b_h);
+    // transform the cropped bbox region
+    transform_->Compute(crop_out, &res_out);
+    // place the transformed region back in the restored input
+    std::shared_ptr<CVTensor> res_img = CVTensor::AsCVTensor(res_out);
+    // check if transformed crop is out of bounds of the box
+    if (res_img->mat().cols > b_w || res_img->mat().rows > b_h || res_img->mat().cols < b_w ||
+        res_img->mat().rows < b_h) {
+      // if so, resize to fit in the box
+      std::shared_ptr<TensorOp> resize_op = std::make_shared<ResizeOp>(b_h, b_w);
+      resize_op->Compute(std::static_pointer_cast<Tensor>(res_img), &res_out);
+      res_img = CVTensor::AsCVTensor(res_out);
+    }
+    res_img->mat().copyTo(input_restore->mat()(cv::Rect(min_x, min_y, res_img->mat().cols, res_img->mat().rows)));
+  }
+  (*output).push_back(std::move(std::static_pointer_cast<Tensor>(input_restore)));
+  (*output).push_back(input[1]);
+  return Status::OK();
+}
+
+}  // namespace dataset
+}  // namespace mindspore

mindspore/ccsrc/dataset/kernels/image/bounding_box_augment_op.h

@@ -0,0 +1,59 @@
+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef DATASET_KERNELS_IMAGE_BOUNDING_BOX_AUGMENT_OP_H_
+#define DATASET_KERNELS_IMAGE_BOUNDING_BOX_AUGMENT_OP_H_
+
+#include <memory>
+#include <random>
+#include <cstdlib>
+#include <opencv2/imgproc/imgproc.hpp>
+#include "dataset/core/tensor.h"
+#include "dataset/kernels/tensor_op.h"
+#include "dataset/util/status.h"
+
+namespace mindspore {
+namespace dataset {
+class BoundingBoxAugOp : public TensorOp {
+ public:
+  // Default values, also used by python_bindings.cc
+  static const float defRatio;
+
+  // Constructor for BoundingBoxAugmentOp
+  // @param std::shared_ptr<TensorOp> transform transform: C++ opration to apply on select bounding boxes
+  // @param float ratio: ratio of bounding boxes to have the transform applied on
+  BoundingBoxAugOp(std::shared_ptr<TensorOp> transform, float ratio);
+
+  ~BoundingBoxAugOp() override = default;
+
+  // Provide stream operator for displaying it
+  friend std::ostream &operator<<(std::ostream &out, const BoundingBoxAugOp &so) {
+    so.Print(out);
+    return out;
+  }
+
+  void Print(std::ostream &out) const override { out << "BoundingBoxAugOp"; }
+
+  Status Compute(const TensorRow &input, TensorRow *output) override;
+
+ private:
+  float ratio_;
+  std::shared_ptr<TensorOp> transform_;
+};
+}  // namespace dataset
+}  // namespace mindspore
+
+#endif  // DATASET_KERNELS_IMAGE_BOUNDING_BOX_AUGMENT_OP_H_

mindspore/ccsrc/dataset/kernels/image/random_horizontal_flip_bbox_op.cc

@@ -0,0 +1,61 @@
+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#include <utility>
+#include "dataset/kernels/image/random_horizontal_flip_bbox_op.h"
+#include "dataset/kernels/image/image_utils.h"
+#include "dataset/util/status.h"
+#include "dataset/core/cv_tensor.h"
+#include "dataset/core/pybind_support.h"
+
+namespace mindspore {
+namespace dataset {
+const float RandomHorizontalFlipWithBBoxOp::kDefProbability = 0.5;
+
+Status RandomHorizontalFlipWithBBoxOp::Compute(const TensorRow &input, TensorRow *output) {
+  IO_CHECK_VECTOR(input, output);
+  BOUNDING_BOX_CHECK(input);
+  if (distribution_(rnd_)) {
+    // To test bounding boxes algorithm, create random bboxes from image dims
+    size_t numOfBBoxes = input[1]->shape()[0];     // set to give number of bboxes
+    float imgCenter = (input[0]->shape()[1] / 2);  // get the center of the image
+
+    for (int i = 0; i < numOfBBoxes; i++) {
+      uint32_t b_w = 0;  // bounding box width
+      uint32_t min_x = 0;
+      // get the required items
+      input[1]->GetItemAt<uint32_t>(&min_x, {i, 0});
+      input[1]->GetItemAt<uint32_t>(&b_w, {i, 2});
+      // do the flip
+      float diff = imgCenter - min_x;          // get distance from min_x to center
+      uint32_t refl_min_x = diff + imgCenter;  // get reflection of min_x
+      uint32_t new_min_x = refl_min_x - b_w;   // subtract from the reflected min_x to get the new one
+
+      input[1]->SetItemAt<uint32_t>({i, 0}, new_min_x);
+    }
+    (*output).push_back(nullptr);
+    (*output).push_back(nullptr);
+    // move input to output pointer of bounding boxes
+    (*output)[1] = std::move(input[1]);
+    // perform HorizontalFlip on the image
+    std::shared_ptr<CVTensor> input_cv = CVTensor::AsCVTensor(std::move(input[0]));
+    return HorizontalFlip(std::static_pointer_cast<Tensor>(input_cv), &(*output)[0]);
+  }
+  *output = input;
+  return Status::OK();
+}
+
+}  // namespace dataset
+}  // namespace mindspore

mindspore/ccsrc/dataset/kernels/image/random_horizontal_flip_bbox_op.h

@@ -0,0 +1,62 @@
+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#ifndef DATASET_KERNELS_IMAGE_RANDOM_HORIZONTAL_FLIP_BBOX_OP_H_
+#define DATASET_KERNELS_IMAGE_RANDOM_HORIZONTAL_FLIP_BBOX_OP_H_
+
+#include <pybind11/numpy.h>
+#include <pybind11/stl.h>
+#include <memory>
+#include <random>
+#include <cstdlib>
+#include <opencv2/imgproc/imgproc.hpp>
+#include "dataset/core/tensor.h"
+#include "dataset/kernels/tensor_op.h"
+#include "dataset/util/random.h"
+#include "dataset/util/status.h"
+#include "pybind11/pybind11.h"
+#include "pybind11/stl_bind.h"
+
+namespace mindspore {
+namespace dataset {
+class RandomHorizontalFlipWithBBoxOp : public TensorOp {
+ public:
+  // Default values, also used by python_bindings.cc
+  static const float kDefProbability;
+
+  explicit RandomHorizontalFlipWithBBoxOp(float probability = kDefProbability) : distribution_(probability) {
+    rnd_.seed(GetSeed());
+  }
+
+  ~RandomHorizontalFlipWithBBoxOp() override = default;
+
+  // Provide stream operator for displaying it
+  friend std::ostream &operator<<(std::ostream &out, const RandomHorizontalFlipWithBBoxOp &so) {
+    so.Print(out);
+    return out;
+  }
+
+  void Print(std::ostream &out) const override { out << "RandomHorizontalFlipWithBBoxOp"; }
+
+  Status Compute(const TensorRow &input, TensorRow *output) override;
+
+ private:
+  std::mt19937 rnd_;
+  std::bernoulli_distribution distribution_;
+};
+}  // namespace dataset
+}  // namespace mindspore
+
+#endif  // DATASET_KERNELS_IMAGE_RANDOM_HORIZONTAL_FLIP_BBOX_OP_H_

mindspore/ccsrc/dataset/kernels/tensor_op.h

@@ -43,6 +43,36 @@
     }                                               \
   } while (false)
 
+#define BOUNDING_BOX_CHECK(input)                                                           \
+  do {                                                                                      \
+    uint32_t num_of_features = input[1]->shape()[1];                                        \
+    if (num_of_features < 4) {                                                              \
+      return Status(StatusCode::kBoundingBoxInvalidShape, __LINE__, __FILE__,               \
+                    "Bounding boxes should be have at least 4 features");                   \
+    }                                                                                       \
+    uint32_t num_of_boxes = input[1]->shape()[0];                                           \
+    uint32_t img_h = input[0]->shape()[0];                                                  \
+    uint32_t img_w = input[0]->shape()[1];                                                  \
+    for (uint32_t i = 0; i < num_of_boxes; i++) {                                           \
+      uint32_t min_x = 0;                                                                   \
+      uint32_t min_y = 0;                                                                   \
+      uint32_t b_w = 0;                                                                     \
+      uint32_t b_h = 0;                                                                     \
+      input[1]->GetItemAt<uint32_t>(&min_x, {i, 0});                                        \
+      input[1]->GetItemAt<uint32_t>(&min_y, {i, 1});                                        \
+      input[1]->GetItemAt<uint32_t>(&b_w, {i, 2});                                          \
+      input[1]->GetItemAt<uint32_t>(&b_h, {i, 3});                                          \
+      if ((min_x + b_w > img_w) || (min_y + b_h > img_h)) {                                 \
+        return Status(StatusCode::kBoundingBoxOutOfBounds, __LINE__, __FILE__,              \
+                      "At least one of the bounding boxes is out of bounds of the image."); \
+      }                                                                                     \
+      if (static_cast<int>(min_x) < 0 || static_cast<int>(min_y) < 0) {                     \
+        return Status(StatusCode::kBoundingBoxOutOfBounds, __LINE__, __FILE__,              \
+                      "At least one of the bounding boxes has negative min_x or min_y.");   \
+      }                                                                                     \
+    }                                                                                       \
+  } while (false)
+
 namespace mindspore {
 namespace dataset {
 // A class that does a computation on  a Tensor

mindspore/ccsrc/dataset/util/status.h

@@ -71,6 +71,8 @@ enum class StatusCode : char {
   kTDTPushFailure = 8,
   kFileNotExist = 9,
   kProfilingError = 10,
+  kBoundingBoxOutOfBounds = 11,
+  kBoundingBoxInvalidShape = 12,
   // Make this error code the last one. Add new error code above it.
   kUnexpectedError = 127
 };

mindspore/dataset/transforms/vision/c_transforms.py

@@ -45,7 +45,7 @@ import mindspore._c_dataengine as cde
 from .utils import Inter, Border
 from .validators import check_prob, check_crop, check_resize_interpolation, check_random_resize_crop, \
     check_normalize_c, check_random_crop, check_random_color_adjust, check_random_rotation, \
-    check_resize, check_rescale, check_pad, check_cutout, check_uniform_augment_cpp
+    check_resize, check_rescale, check_pad, check_cutout, check_uniform_augment_cpp, check_bounding_box_augment_cpp
 
 DE_C_INTER_MODE = {Inter.NEAREST: cde.InterpolationMode.DE_INTER_NEAREST_NEIGHBOUR,
                    Inter.LINEAR: cde.InterpolationMode.DE_INTER_LINEAR,
@@ -163,6 +163,21 @@ class RandomHorizontalFlip(cde.RandomHorizontalFlipOp):
         super().__init__(prob)
 
 
+class RandomHorizontalFlipWithBBox(cde.RandomHorizontalFlipWithBBoxOp):
+    """
+    Flip the input image horizontally, randomly with a given probability.
+    Maintains data integrity by also flipping bounding boxes in an object detection pipeline.
+
+    Args:
+        prob (float): Probability of the image being flipped (default=0.5).
+    """
+
+    @check_prob
+    def __init__(self, prob=0.5):
+        self.prob = prob
+        super().__init__(prob)
+
+
 class RandomVerticalFlip(cde.RandomVerticalFlipOp):
     """
     Flip the input image vertically, randomly with a given probability.
@@ -177,6 +192,21 @@ class RandomVerticalFlip(cde.RandomVerticalFlipOp):
         super().__init__(prob)
 
 
+class BoundingBoxAug(cde.BoundingBoxAugOp):
+    """
+    Flip the input image vertically, randomly with a given probability.
+
+    Args:
+        transform: C++ operation (python OPs are not accepted).
+        ratio (float): Ratio of bounding boxes to apply augmentation on. Range: [0,1] (default=1).
+    """
+    @check_bounding_box_augment_cpp
+    def __init__(self, transform, ratio=0.3):
+        self.ratio = ratio
+        self.transform = transform
+        super().__init__(transform, ratio)
+
+
 class Resize(cde.ResizeOp):
     """
     Resize the input image to the given size.

mindspore/dataset/transforms/vision/validators.py

@@ -852,6 +852,32 @@ def check_uniform_augment_cpp(method):
     return new_method
 
 
+def check_bounding_box_augment_cpp(method):
+    """Wrapper method to check the parameters of BoundingBoxAugment cpp op."""
+
+    @wraps(method)
+    def new_method(self, *args, **kwargs):
+        transform, ratio = (list(args) + 2 * [None])[:2]
+        if "transform" in kwargs:
+            transform = kwargs.get("transform")
+        if "ratio" in kwargs:
+            ratio = kwargs.get("ratio")
+        if ratio is not None:
+            check_value(ratio, [0., 1.])
+            kwargs["ratio"] = ratio
+        else:
+            ratio = 0.3
+        if not isinstance(ratio, float) and not isinstance(ratio, int):
+            raise ValueError("Ratio should be an int or float.")
+        if not isinstance(transform, TensorOp):
+            raise ValueError("Transform can only be a C++ operation.")
+        kwargs["transform"] = transform
+        kwargs["ratio"] = ratio
+        return method(self, **kwargs)
+
+    return new_method
+
+
 def check_uniform_augment_py(method):
     """Wrapper method to check the parameters of python UniformAugment op."""
 

tests/ut/data/dataset/testVOC2012_2/Annotations/121.xml

@@ -0,0 +1,27 @@
+<annotation>
+	<folder>VOC2012</folder>
+	<filename>121.jpg</filename>
+	<source>
+		<database>simulate VOC2007 Database</database>
+		<annotation>simulate VOC2007</annotation>
+		<image>flickr</image>
+	</source>
+	<size>
+		<width>500</width>
+		<height>375</height>
+		<depth>3</depth>
+	</size>
+	<segmented>1</segmented>
+	<object>
+		<name>dog</name>
+		<pose>Frontal</pose>
+		<truncated>0</truncated>
+		<difficult>0</difficult>
+		<bndbox>
+			<xmin>55</xmin>
+			<ymin>34</ymin>
+			<xmax>624</xmax>
+			<ymax>555</ymax>
+		</bndbox>
+	</object>
+</annotation>

tests/ut/data/dataset/testVOC2012_2/Annotations/123.xml

@@ -0,0 +1,27 @@
+<annotation>
+	<folder>VOC2012</folder>
+	<filename>123.jpg</filename>
+	<source>
+		<database>simulate VOC2007 Database</database>
+		<annotation>simulate VOC2007</annotation>
+		<image>flickr</image>
+	</source>
+	<size>
+		<width>500</width>
+		<height>375</height>
+		<depth>3</depth>
+	</size>
+	<segmented>1</segmented>
+	<object>
+		<name>car</name>
+		<pose>Unspecified</pose>
+		<truncated>1</truncated>
+		<difficult>0</difficult>
+		<bndbox>
+			<xmin>42</xmin>
+			<ymin>6</ymin>
+			<xmax>610</xmax>
+			<ymax>600</ymax>
+		</bndbox>
+	</object>
+</annotation>

tests/ut/data/dataset/testVOC2012_2/Annotations/129.xml

@@ -0,0 +1,27 @@
+<annotation>
+	<folder>VOC2012</folder>
+	<filename>129.jpg</filename>
+	<source>
+		<database>simulate VOC2007 Database</database>
+		<annotation>simulate VOC2007</annotation>
+		<image>flickr</image>
+	</source>
+	<size>
+		<width>500</width>
+		<height>375</height>
+		<depth>3</depth>
+	</size>
+	<segmented>1</segmented>
+	<object>
+		<name>dog</name>
+		<pose>Frontal</pose>
+		<truncated>0</truncated>
+		<difficult>0</difficult>
+		<bndbox>
+			<xmin>1328</xmin>
+			<ymin>431</ymin>
+			<xmax>2662</xmax>
+			<ymax>1695</ymax>
+		</bndbox>
+	</object>
+</annotation>

tests/ut/data/dataset/testVOC2012_2/Annotations/15.xml

@@ -0,0 +1,27 @@
+<annotation>
+	<folder>VOC2012</folder>
+	<filename>32.jpg</filename>
+	<source>
+		<database>simulate VOC2007 Database</database>
+		<annotation>simulate VOC2007</annotation>
+		<image>flickr</image>
+	</source>
+	<size>
+		<width>500</width>
+		<height>281</height>
+		<depth>3</depth>
+	</size>
+	<segmented>1</segmented>
+	<object>
+		<name>train</name>
+		<pose>Frontal</pose>
+		<truncated>0</truncated>
+		<difficult>0</difficult>
+		<bndbox>
+			<xmin>1168</xmin>
+			<ymin>405</ymin>
+			<xmax>3270</xmax>
+			<ymax>2022</ymax>
+		</bndbox>
+	</object>
+</annotation>

tests/ut/data/dataset/testVOC2012_2/Annotations/32.xml

@@ -0,0 +1,27 @@
+<annotation>
+	<folder>VOC2012</folder>
+	<filename>32.jpg</filename>
+	<source>
+		<database>simulate VOC2007 Database</database>
+		<annotation>simulate VOC2007</annotation>
+		<image>flickr</image>
+	</source>
+	<size>
+		<width>500</width>
+		<height>281</height>
+		<depth>3</depth>
+	</size>
+	<segmented>1</segmented>
+	<object>
+		<name>train</name>
+		<pose>Frontal</pose>
+		<truncated>0</truncated>
+		<difficult>0</difficult>
+		<bndbox>
+			<xmin>1168</xmin>
+			<ymin>405</ymin>
+			<xmax>3270</xmax>
+			<ymax>2022</ymax>
+		</bndbox>
+	</object>
+	</annotation>

tests/ut/data/dataset/testVOC2012_2/Annotations/33.xml

@@ -0,0 +1,27 @@
+<annotation>
+	<folder>VOC2012</folder>
+	<filename>33.jpg</filename>
+	<source>
+		<database>simulate VOC2007 Database</database>
+		<annotation>simulate VOC2007</annotation>
+		<image>flickr</image>
+	</source>
+	<size>
+		<width>500</width>
+		<height>366</height>
+		<depth>3</depth>
+	</size>
+	<segmented>1</segmented>
+	<object>
+		<name>person</name>
+		<pose>Unspecified</pose>
+		<truncated>0</truncated>
+		<difficult>0</difficult>
+		<bndbox>
+			<xmin>1168</xmin>
+			<ymin>395</ymin>
+			<xmax>2859</xmax>
+			<ymax>2084</ymax>
+		</bndbox>
+	</object>
+	</annotation>

tests/ut/data/dataset/testVOC2012_2/Annotations/39.xml

@@ -0,0 +1,27 @@
+<annotation>
+	<folder>VOC2012</folder>
+	<filename>39.jpg</filename>
+	<source>
+		<database>simulate VOC2007 Database</database>
+		<annotation>simulate VOC2007</annotation>
+		<image>flickr</image>
+	</source>
+	<size>
+		<width>500</width>
+		<height>375</height>
+		<depth>3</depth>
+	</size>
+	<segmented>1</segmented>
+	<object>
+		<name>dog</name>
+		<pose>Unspecified</pose>
+		<truncated>0</truncated>
+		<difficult>0</difficult>
+		<bndbox>
+			<xmin>684</xmin>
+			<ymin>311</ymin>
+			<xmax>3112</xmax>
+			<ymax>1820</ymax>
+		</bndbox>
+	</object>
+</annotation>

tests/ut/data/dataset/testVOC2012_2/Annotations/42.xml

@@ -0,0 +1,27 @@
+<annotation>
+	<folder>VOC2012</folder>
+	<filename>42.jpg</filename>
+	<source>
+		<database>simulate VOC2007 Database</database>
+		<annotation>simulate VOC2007</annotation>
+		<image>flickr</image>
+	</source>
+	<size>
+		<width>500</width>
+		<height>335</height>
+		<depth>3</depth>
+	</size>
+	<segmented>1</segmented>
+	<object>
+		<name>person</name>
+		<pose>Unspecified</pose>
+		<truncated>1</truncated>
+		<difficult>0</difficult>
+		<bndbox>
+			<xmin>874</xmin>
+			<ymin>152</ymin>
+			<xmax>2827</xmax>
+			<ymax>2000</ymax>
+		</bndbox>
+	</object>
+</annotation>

tests/ut/data/dataset/testVOC2012_2/Annotations/61.xml

@@ -0,0 +1,39 @@
+<annotation>
+	<folder>VOC2012</folder>
+	<filename>61.jpg</filename>
+	<source>
+		<database>simulate VOC2007 Database</database>
+		<annotation>simulate VOC2007</annotation>
+		<image>flickr</image>
+	</source>
+	<size>
+		<width>500</width>
+		<height>333</height>
+		<depth>3</depth>
+	</size>
+	<segmented>1</segmented>
+	<object>
+		<name>train</name>
+		<pose>Unspecified</pose>
+		<truncated>0</truncated>
+		<difficult>0</difficult>
+		<bndbox>
+			<xmin>25</xmin>
+			<ymin>40</ymin>
+			<xmax>641</xmax>
+			<ymax>613</ymax>
+		</bndbox>
+	</object>
+	<object>
+		<name>person</name>
+		<pose>Frontal</pose>
+		<truncated>0</truncated>
+		<difficult>0</difficult>
+		<bndbox>
+			<xmin>204</xmin>
+			<ymin>198</ymin>
+			<xmax>271</xmax>
+			<ymax>293</ymax>
+		</bndbox>
+	</object>
+</annotation>

tests/ut/data/dataset/testVOC2012_2/Annotations/63.xml

@@ -0,0 +1,39 @@
+<annotation>
+	<folder>VOC2012</folder>
+	<filename>63.jpg</filename>
+	<source>
+		<database>simulate VOC2007 Database</database>
+		<annotation>simulate VOC2007</annotation>
+		<image>flickr</image>
+	</source>
+	<size>
+		<width>500</width>
+		<height>375</height>
+		<depth>3</depth>
+	</size>
+	<segmented>1</segmented>
+	<object>
+		<name>cat</name>
+		<pose>Unspecified</pose>
+		<truncated>0</truncated>
+		<difficult>0</difficult>
+		<bndbox>
+			<xmin>23</xmin>
+			<ymin>17</ymin>
+			<xmax>565</xmax>
+			<ymax>591</ymax>
+		</bndbox>
+	</object>
+	<object>
+		<name>chair</name>
+		<pose>Frontal</pose>
+		<truncated>1</truncated>
+		<difficult>0</difficult>
+		<bndbox>
+			<xmin>36</xmin>
+			<ymin>11</ymin>
+			<xmax>439</xmax>
+			<ymax>499</ymax>
+		</bndbox>
+	</object>
+</annotation>

tests/ut/data/dataset/testVOC2012_2/Annotations/68.xml

@@ -0,0 +1,27 @@
+<annotation>
+	<folder>VOC2012</folder>
+	<filename>68.jpg</filename>
+	<source>
+		<database>simulate VOC2007 Database</database>
+		<annotation>simulate VOC2007</annotation>
+		<image>flickr</image>
+	</source>
+	<size>
+		<width>500</width>
+		<height>375</height>
+		<depth>3</depth>
+	</size>
+	<segmented>1</segmented>
+	<object>
+		<name>cat</name>
+		<pose>Unspecified</pose>
+		<truncated>1</truncated>
+		<difficult>0</difficult>
+		<bndbox>
+			<xmin>35</xmin>
+			<ymin>11</ymin>
+			<xmax>564</xmax>
+			<ymax>545</ymax>
+		</bndbox>
+	</object>
+</annotation>

tests/ut/data/dataset/testVOC2012_2/Annotations/invalidxml.xml

@@ -0,0 +1 @@
+invalidxml

tests/ut/data/dataset/testVOC2012_2/Annotations/xmlnoobject.xml

@@ -0,0 +1,15 @@
+<annotation>
+	<folder>VOC2012</folder>
+	<filename>33.jpg</filename>
+	<source>
+		<database>simulate VOC2007 Database</database>
+		<annotation>simulate VOC2007</annotation>
+		<image>flickr</image>
+	</source>
+	<size>
+		<width>500</width>
+		<height>366</height>
+		<depth>3</depth>
+	</size>
+	<segmented>1</segmented>
+</annotation>

tests/ut/data/dataset/testVOC2012_2/ImageSets/Main/invalidxml.txt

@@ -0,0 +1 @@
+invalidxml

tests/ut/data/dataset/testVOC2012_2/ImageSets/Main/train.txt

@@ -0,0 +1,11 @@
+15
+32
+33
+39
+42
+61
+63
+68
+121
+123
+129