!2920 Updating 3 BBox AugOps for Float + Common Files for BBox Float Support -...

!2920 Updating 3 BBox AugOps for Float + Common Files for BBox Float Support - (VertFlip/RandomCrop/RandomResizeCrop)
Merge pull request !2920 from danishnxt/AugOps-M1-Python

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

@@ -740,22 +740,16 @@ Status UpdateBBoxesForCrop(std::shared_ptr<Tensor> *bboxList, size_t *bboxCount,
                            int CB_Ymax) {
   // PASS LIST, COUNT OF BOUNDING BOXES
   // Also PAss X/Y Min/Max of image cropped region - normally obtained from 'GetCropBox' functions
-  uint32_t bb_Xmin_t, bb_Ymin_t, bb_Xmax_t, bb_Ymax_t;
-
+  float bb_Xmin = 0.0, bb_Ymin = 0.0, bb_Xmax = 0.0, bb_Ymax = 0.0;
   std::vector<int> correct_ind;
-  std::vector<uint32_t> copyVals;
+  std::vector<float> copyVals;
   dsize_t bboxDim = (*bboxList)->shape()[1];
   bool retFlag = false;  // true unless overlap found
   for (int i = 0; i < *bboxCount; i++) {
-    int bb_Xmin, bb_Xmax, bb_Ymin, bb_Ymax;
-    RETURN_IF_NOT_OK((*bboxList)->GetUnsignedIntAt(&bb_Xmin_t, {i, 0}));
-    RETURN_IF_NOT_OK((*bboxList)->GetUnsignedIntAt(&bb_Ymin_t, {i, 1}));
-    RETURN_IF_NOT_OK((*bboxList)->GetUnsignedIntAt(&bb_Xmax_t, {i, 2}));
-    RETURN_IF_NOT_OK((*bboxList)->GetUnsignedIntAt(&bb_Ymax_t, {i, 3}));
-    bb_Xmin = bb_Xmin_t;
-    bb_Ymin = bb_Ymin_t;
-    bb_Xmax = bb_Xmax_t;
-    bb_Ymax = bb_Ymax_t;
+    RETURN_IF_NOT_OK((*bboxList)->GetItemAt<float>(&bb_Xmin, {i, 0}));
+    RETURN_IF_NOT_OK((*bboxList)->GetItemAt<float>(&bb_Ymin, {i, 1}));
+    RETURN_IF_NOT_OK((*bboxList)->GetItemAt<float>(&bb_Xmax, {i, 2}));
+    RETURN_IF_NOT_OK((*bboxList)->GetItemAt<float>(&bb_Ymax, {i, 3}));
     bb_Xmax = bb_Xmin + bb_Xmax;
     bb_Ymax = bb_Ymin + bb_Ymax;
     // check for image / BB overlap
@@ -766,23 +760,23 @@ Status UpdateBBoxesForCrop(std::shared_ptr<Tensor> *bboxList, size_t *bboxCount,
     correct_ind.push_back(i);
     // adjust BBox corners by bringing into new CropBox if beyond
     // Also reseting/adjusting for boxes to lie within CropBox instead of Image - subtract CropBox Xmin/YMin
-    bb_Xmin = bb_Xmin - (std::min(0, (bb_Xmin - CB_Xmin)) + CB_Xmin);
-    bb_Xmax = bb_Xmax - (std::max(0, (bb_Xmax - CB_Xmax)) + CB_Xmin);
-    bb_Ymin = bb_Ymin - (std::min(0, (bb_Ymin - CB_Ymin)) + CB_Ymin);
-    bb_Ymax = bb_Ymax - (std::max(0, (bb_Ymax - CB_Ymax)) + CB_Ymin);
+    bb_Xmin = bb_Xmin - (std::min(static_cast<float>(0.0), (bb_Xmin - CB_Xmin)) + CB_Xmin);
+    bb_Xmax = bb_Xmax - (std::max(static_cast<float>(0.0), (bb_Xmax - CB_Xmax)) + CB_Xmin);
+    bb_Ymin = bb_Ymin - (std::min(static_cast<float>(0.0), (bb_Ymin - CB_Ymin)) + CB_Ymin);
+    bb_Ymax = bb_Ymax - (std::max(static_cast<float>(0.0), (bb_Ymax - CB_Ymax)) + CB_Ymin);
     // reset min values and calculate width/height from Box corners
-    RETURN_IF_NOT_OK((*bboxList)->SetItemAt({i, 0}, static_cast<uint32_t>(bb_Xmin)));
-    RETURN_IF_NOT_OK((*bboxList)->SetItemAt({i, 1}, static_cast<uint32_t>(bb_Ymin)));
-    RETURN_IF_NOT_OK((*bboxList)->SetItemAt({i, 2}, static_cast<uint32_t>(bb_Xmax - bb_Xmin)));
-    RETURN_IF_NOT_OK((*bboxList)->SetItemAt({i, 3}, static_cast<uint32_t>(bb_Ymax - bb_Ymin)));
+    RETURN_IF_NOT_OK((*bboxList)->SetItemAt({i, 0}, bb_Xmin));
+    RETURN_IF_NOT_OK((*bboxList)->SetItemAt({i, 1}, bb_Ymin));
+    RETURN_IF_NOT_OK((*bboxList)->SetItemAt({i, 2}, bb_Xmax - bb_Xmin));
+    RETURN_IF_NOT_OK((*bboxList)->SetItemAt({i, 3}, bb_Ymax - bb_Ymin));
   }
   // create new tensor and copy over bboxes still valid to the image
   // bboxes outside of new cropped region are ignored - empty tensor returned in case of none
   *bboxCount = correct_ind.size();
-  uint32_t temp;
+  float temp = 0.0;
   for (auto slice : correct_ind) {  // for every index in the loop
     for (int ix = 0; ix < bboxDim; ix++) {
-      RETURN_IF_NOT_OK((*bboxList)->GetUnsignedIntAt(&temp, {slice, ix}));
+      RETURN_IF_NOT_OK((*bboxList)->GetItemAt<float>(&temp, {slice, ix}));
       copyVals.push_back(temp);
     }
   }
@@ -794,11 +788,11 @@ Status UpdateBBoxesForCrop(std::shared_ptr<Tensor> *bboxList, size_t *bboxCount,
 
 Status PadBBoxes(const std::shared_ptr<Tensor> *bboxList, const size_t &bboxCount, int32_t pad_top, int32_t pad_left) {
   for (int i = 0; i < bboxCount; i++) {
-    uint32_t xMin, yMin;
-    RETURN_IF_NOT_OK((*bboxList)->GetUnsignedIntAt(&xMin, {i, 0}));
-    RETURN_IF_NOT_OK((*bboxList)->GetUnsignedIntAt(&yMin, {i, 1}));
-    xMin += static_cast<uint32_t>(pad_left);  // should not be negative
-    yMin += static_cast<uint32_t>(pad_top);
+    float xMin = 0.0, yMin = 0.0;
+    RETURN_IF_NOT_OK((*bboxList)->GetItemAt<float>(&xMin, {i, 0}));
+    RETURN_IF_NOT_OK((*bboxList)->GetItemAt<float>(&yMin, {i, 1}));
+    xMin += pad_left;
+    yMin += pad_top;
     RETURN_IF_NOT_OK((*bboxList)->SetItemAt({i, 0}, xMin));
     RETURN_IF_NOT_OK((*bboxList)->SetItemAt({i, 1}, yMin));
   }
@@ -807,16 +801,16 @@ Status PadBBoxes(const std::shared_ptr<Tensor> *bboxList, const size_t &bboxCoun
 
 Status UpdateBBoxesForResize(const std::shared_ptr<Tensor> &bboxList, const size_t &bboxCount, int32_t target_width_,
                              int32_t target_height_, int orig_width, int orig_height) {
-  uint32_t bb_Xmin, bb_Ymin, bb_Xwidth, bb_Ywidth;
-  // cast to float to preseve fractional
-  double W_aspRatio = (target_width_ * 1.0) / (orig_width * 1.0);
-  double H_aspRatio = (target_height_ * 1.0) / (orig_height * 1.0);
+  float bb_Xmin = 0, bb_Ymin = 0, bb_Xwidth = 0, bb_Ywidth = 0;
+  // cast to float to preserve fractional
+  float W_aspRatio = (target_width_ * 1.0) / (orig_width * 1.0);
+  float H_aspRatio = (target_height_ * 1.0) / (orig_height * 1.0);
   for (int i = 0; i < bboxCount; i++) {
     // for each bounding box
-    RETURN_IF_NOT_OK(bboxList->GetUnsignedIntAt(&bb_Xmin, {i, 0}));
-    RETURN_IF_NOT_OK(bboxList->GetUnsignedIntAt(&bb_Ymin, {i, 1}));
-    RETURN_IF_NOT_OK(bboxList->GetUnsignedIntAt(&bb_Xwidth, {i, 2}));
-    RETURN_IF_NOT_OK(bboxList->GetUnsignedIntAt(&bb_Ywidth, {i, 3}));
+    RETURN_IF_NOT_OK(bboxList->GetItemAt<float>(&bb_Xmin, {i, 0}));
+    RETURN_IF_NOT_OK(bboxList->GetItemAt<float>(&bb_Ymin, {i, 1}));
+    RETURN_IF_NOT_OK(bboxList->GetItemAt<float>(&bb_Xwidth, {i, 2}));
+    RETURN_IF_NOT_OK(bboxList->GetItemAt<float>(&bb_Ywidth, {i, 3}));
     // update positions and widths
     bb_Xmin = bb_Xmin * W_aspRatio;
     bb_Ymin = bb_Ymin * H_aspRatio;

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

@@ -34,14 +34,13 @@ Status RandomVerticalFlipWithBBoxOp::Compute(const TensorRow &input, TensorRow *
     // one time allocation -> updated in the loop
     // type defined based on VOC test dataset
     for (int i = 0; i < boxCount; i++) {
-      uint32_t boxCorner_y = 0;
-      uint32_t boxHeight = 0;
-      uint32_t newBoxCorner_y = 0;
-      RETURN_IF_NOT_OK(input[1]->GetUnsignedIntAt(&boxCorner_y, {i, 1}));  // get min y of bbox
-      RETURN_IF_NOT_OK(input[1]->GetUnsignedIntAt(&boxHeight, {i, 3}));    // get height of bbox
+      float boxCorner_y = 0.0, boxHeight = 0.0;
+      float newBoxCorner_y = 0.0;
+      RETURN_IF_NOT_OK(input[1]->GetItemAt<float>(&boxCorner_y, {i, 1}));  // get min y of bbox
+      RETURN_IF_NOT_OK(input[1]->GetItemAt<float>(&boxHeight, {i, 3}));    // get height of bbox
 
       // subtract (curCorner + height) from (max) for new Corner position
-      newBoxCorner_y = (imHeight - 1) - ((boxCorner_y + boxHeight) - 1);
+      newBoxCorner_y = (imHeight - 1.0) - ((boxCorner_y + boxHeight) - 1.0);
       RETURN_IF_NOT_OK(input[1]->SetItemAt({i, 1}, newBoxCorner_y));
     }
 

mindspore/ccsrc/dataset/kernels/tensor_op.h

@@ -62,14 +62,16 @@
     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});                                          \
+      float min_x = 0.0, min_y = 0.0, b_w = 0.0, b_h = 0.0;                                 \
+      bool passing_data_fetch = true;                                                       \
+      passing_data_fetch &= input[1]->GetItemAt<float>(&min_x, {i, 0}).IsOk();              \
+      passing_data_fetch &= input[1]->GetItemAt<float>(&min_y, {i, 1}).IsOk();              \
+      passing_data_fetch &= input[1]->GetItemAt<float>(&b_w, {i, 2}).IsOk();                \
+      passing_data_fetch &= input[1]->GetItemAt<float>(&b_h, {i, 3}).IsOk();                \
+      if (!passing_data_fetch) {                                                            \
+        return Status(StatusCode::kUnexpectedError, __LINE__, __FILE__,                     \
+                      "Fetching BBox values failed in BOUNDING_BOX_CHECK.");                \
+      }                                                                                     \
       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."); \

tests/ut/cpp/dataset/common/bboxop_common.cc

@@ -118,14 +118,11 @@ void BBoxOpCommon::SaveImagesWithAnnotations(BBoxOpCommon::FileType type, const
     bool passing_data_fetch = true;
     // For each bounding box draw on the image.
     for (uint32_t i = 0; i < num_of_boxes; i++) {
-      uint32_t x = 0;
-      uint32_t y = 0;
-      uint32_t w = 0;
-      uint32_t h = 0;
-      passing_data_fetch &= row[1]->GetUnsignedIntAt(&x, {i, 0}).IsOk();
-      passing_data_fetch &= row[1]->GetUnsignedIntAt(&y, {i, 1}).IsOk();
-      passing_data_fetch &= row[1]->GetUnsignedIntAt(&w, {i, 2}).IsOk();
-      passing_data_fetch &= row[1]->GetUnsignedIntAt(&h, {i, 3}).IsOk();
+      float x = 0.0, y = 0.0, w = 0.0, h = 0.0;
+      passing_data_fetch &= row[1]->GetItemAt<float>(&x, {i, 0}).IsOk();
+      passing_data_fetch &= row[1]->GetItemAt<float>(&y, {i, 1}).IsOk();
+      passing_data_fetch &= row[1]->GetItemAt<float>(&w, {i, 2}).IsOk();
+      passing_data_fetch &= row[1]->GetItemAt<float>(&h, {i, 3}).IsOk();
       if (!passing_data_fetch) {
         MS_LOG(ERROR) << "Fetching bbox coordinates failed in SaveImagesWithAnnotations.";
         EXPECT_TRUE(passing_data_fetch);
@@ -193,24 +190,24 @@ bool BBoxOpCommon::LoadAnnotationFile(const std::string &path, std::shared_ptr<T
     MS_LOG(ERROR) << "No object find in " + path;
     return false;
   }
-  std::vector<uint32_t> return_value_list;
+  std::vector<float> return_value_list;
   dsize_t bbox_count = 0;      // keep track of number of bboxes in file
   dsize_t bbox_val_count = 4;  // creating bboxes of size 4 to test function
   // FILE OK TO READ
   while (object != nullptr) {
     bbox_count += 1;
     std::string label_name;
-    uint32_t xmin = 0, ymin = 0, xmax = 0, ymax = 0;
+    float xmin = 0.0, ymin = 0.0, xmax = 0.0, ymax = 0.0;
     XMLElement *bbox_node = object->FirstChildElement("bndbox");
     if (bbox_node != nullptr) {
       XMLElement *xmin_node = bbox_node->FirstChildElement("xmin");
-      if (xmin_node != nullptr) xmin = xmin_node->UnsignedText();
+      if (xmin_node != nullptr) xmin = xmin_node->FloatText();
       XMLElement *ymin_node = bbox_node->FirstChildElement("ymin");
-      if (ymin_node != nullptr) ymin = ymin_node->UnsignedText();
+      if (ymin_node != nullptr) ymin = ymin_node->FloatText();
       XMLElement *xmax_node = bbox_node->FirstChildElement("xmax");
-      if (xmax_node != nullptr) xmax = xmax_node->UnsignedText();
+      if (xmax_node != nullptr) xmax = xmax_node->FloatText();
       XMLElement *ymax_node = bbox_node->FirstChildElement("ymax");
-      if (ymax_node != nullptr) ymax = ymax_node->UnsignedText();
+      if (ymax_node != nullptr) ymax = ymax_node->FloatText();
     } else {
       MS_LOG(ERROR) << "bndbox dismatch in " + path;
       return false;

tests/ut/python/dataset/test_random_crop_and_resize_with_bbox.py

+# 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.
+# ==============================================================================
+"""
+Testing RandomCropAndResizeWithBBox op in DE
+"""
+import numpy as np
+import mindspore.dataset as ds
+import mindspore.dataset.transforms.vision.c_transforms as c_vision
+
+from mindspore import log as logger
+from util import visualize_with_bounding_boxes, InvalidBBoxType, check_bad_bbox, \
+    config_get_set_seed, config_get_set_num_parallel_workers, save_and_check_md5
+
+GENERATE_GOLDEN = False
+
+# Updated VOC dataset with correct annotations - DATA_DIR
+DATA_DIR_VOC = "../data/dataset/testVOC2012_2"
+# COCO dataset - DATA_DIR, ANNOTATION_DIR
+DATA_DIR_COCO = ["../data/dataset/testCOCO/train/", "../data/dataset/testCOCO/annotations/train.json"]
+
+
+def test_random_resized_crop_with_bbox_op_c(plot_vis=False):
+    """
+    Prints images and bboxes side by side with and without RandomResizedCropWithBBox Op applied,
+    tests with MD5 check, expected to pass
+    """
+    logger.info("test_random_resized_crop_with_bbox_op_c")
+
+    original_seed = config_get_set_seed(23415)
+    original_num_parallel_workers = config_get_set_num_parallel_workers(1)
+
+    # Load dataset
+    dataVoc1 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False)
+    dataVoc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False)
+
+    test_op = c_vision.RandomResizedCropWithBBox((256, 512), (0.5, 0.5), (0.5, 0.5))
+
+    # map to apply ops
+    dataVoc2 = dataVoc2.map(input_columns=["image", "annotation"],
+                            output_columns=["image", "annotation"],
+                            columns_order=["image", "annotation"],
+                            operations=[test_op])
+
+    filename = "random_resized_crop_with_bbox_01_c_result.npz"
+    save_and_check_md5(dataVoc2, filename, generate_golden=GENERATE_GOLDEN)
+
+    unaugSamp, augSamp = [], []
+
+    for unAug, Aug in zip(dataVoc1.create_dict_iterator(), dataVoc2.create_dict_iterator()):
+        unaugSamp.append(unAug)
+        augSamp.append(Aug)
+
+    if plot_vis:
+        visualize_with_bounding_boxes(unaugSamp, augSamp)
+
+    # Restore config setting
+    ds.config.set_seed(original_seed)
+    ds.config.set_num_parallel_workers(original_num_parallel_workers)
+
+
+def test_random_resized_crop_with_bbox_op_coco_c(plot_vis=False):
+    """
+    Prints images and bboxes side by side with and without RandomResizedCropWithBBox Op applied,
+    Testing with Coco dataset
+    """
+    logger.info("test_random_resized_crop_with_bbox_op_coco_c")
+    # load dataset
+    dataCoco1 = ds.CocoDataset(DATA_DIR_COCO[0], annotation_file=DATA_DIR_COCO[1], task="Detection",
+                               decode=True, shuffle=False)
+
+    dataCoco2 = ds.CocoDataset(DATA_DIR_COCO[0], annotation_file=DATA_DIR_COCO[1], task="Detection",
+                               decode=True, shuffle=False)
+
+    test_op = c_vision.RandomResizedCropWithBBox((512, 512), (0.5, 1), (0.5, 1))
+
+    dataCoco2 = dataCoco2.map(input_columns=["image", "bbox"],
+                              output_columns=["image", "bbox"],
+                              columns_order=["image", "bbox"],
+                              operations=[test_op])
+
+    unaugSamp, augSamp = [], []
+
+    for unAug, Aug in zip(dataCoco1.create_dict_iterator(), dataCoco2.create_dict_iterator()):
+        unaugSamp.append(unAug)
+        augSamp.append(Aug)
+
+    if plot_vis:
+        visualize_with_bounding_boxes(unaugSamp, augSamp, "bbox")
+
+
+def test_random_resized_crop_with_bbox_op_edge_c(plot_vis=False):
+    """
+    Prints images and bboxes side by side with and without RandomResizedCropWithBBox Op applied,
+    tests on dynamically generated edge case, expected to pass
+    """
+    logger.info("test_random_resized_crop_with_bbox_op_edge_c")
+
+    # Load dataset
+    dataVoc1 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False)
+    dataVoc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False)
+
+    test_op = c_vision.RandomResizedCropWithBBox((256, 512), (0.5, 0.5), (0.5, 0.5))
+
+    # maps to convert data into valid edge case data
+    dataVoc1 = dataVoc1.map(input_columns=["image", "annotation"],
+                            output_columns=["image", "annotation"],
+                            columns_order=["image", "annotation"],
+                            operations=[lambda img, bboxes: (img, np.array([[0, 0, img.shape[1], img.shape[0]]]).astype(bboxes.dtype))])
+
+    # Test Op added to list of Operations here
+    dataVoc2 = dataVoc2.map(input_columns=["image", "annotation"],
+                            output_columns=["image", "annotation"],
+                            columns_order=["image", "annotation"],
+                            operations=[lambda img, bboxes: (img, np.array([[0, 0, img.shape[1], img.shape[0]]]).astype(bboxes.dtype)), test_op])
+
+    unaugSamp, augSamp = [], []
+
+    for unAug, Aug in zip(dataVoc1.create_dict_iterator(), dataVoc2.create_dict_iterator()):
+        unaugSamp.append(unAug)
+        augSamp.append(Aug)
+
+    if plot_vis:
+        visualize_with_bounding_boxes(unaugSamp, augSamp)
+
+
+def test_random_resized_crop_with_bbox_op_invalid_c():
+    """
+    Tests RandomResizedCropWithBBox on invalid constructor parameters, expected to raise ValueError
+    """
+    logger.info("test_random_resized_crop_with_bbox_op_invalid_c")
+
+    # Load dataset, only Augmented Dataset as test will raise ValueError
+    dataVoc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False)
+
+    try:
+        # If input range of scale is not in the order of (min, max), ValueError will be raised.
+        test_op = c_vision.RandomResizedCropWithBBox((256, 512), (1, 0.5), (0.5, 0.5))
+
+        # map to apply ops
+        dataVoc2 = dataVoc2.map(input_columns=["image", "annotation"],
+                                output_columns=["image", "annotation"],
+                                columns_order=["image", "annotation"],
+                                operations=[test_op])
+
+        for _ in dataVoc2.create_dict_iterator():
+            break
+
+    except ValueError as err:
+        logger.info("Got an exception in DE: {}".format(str(err)))
+        assert "Input range is not valid" in str(err)
+
+
+def test_random_resized_crop_with_bbox_op_invalid2_c():
+    """
+     Tests RandomResizedCropWithBBox Op on invalid constructor parameters, expected to raise ValueError
+    """
+    logger.info("test_random_resized_crop_with_bbox_op_invalid2_c")
+    # Load dataset # only loading the to AugDataset as test will fail on this
+    dataVoc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False)
+
+    try:
+        # If input range of ratio is not in the order of (min, max), ValueError will be raised.
+        test_op = c_vision.RandomResizedCropWithBBox((256, 512), (1, 1), (1, 0.5))
+
+        # map to apply ops
+        dataVoc2 = dataVoc2.map(input_columns=["image", "annotation"],
+                                output_columns=["image", "annotation"],
+                                columns_order=["image", "annotation"],
+                                operations=[test_op])
+
+        for _ in dataVoc2.create_dict_iterator():
+            break
+
+    except ValueError as err:
+        logger.info("Got an exception in DE: {}".format(str(err)))
+        assert "Input range is not valid" in str(err)
+
+
+def test_random_resized_crop_with_bbox_op_bad_c():
+    """
+    Test RandomCropWithBBox op with invalid bounding boxes, expected to catch multiple errors.
+    """
+    logger.info("test_random_resized_crop_with_bbox_op_bad_c")
+    test_op = c_vision.RandomResizedCropWithBBox((256, 512), (0.5, 0.5), (0.5, 0.5))
+
+    data_voc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False)
+    check_bad_bbox(data_voc2, test_op, InvalidBBoxType.WidthOverflow, "bounding boxes is out of bounds of the image")
+    data_voc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False)
+    check_bad_bbox(data_voc2, test_op, InvalidBBoxType.HeightOverflow, "bounding boxes is out of bounds of the image")
+    data_voc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False)
+    check_bad_bbox(data_voc2, test_op, InvalidBBoxType.NegativeXY, "min_x")
+    data_voc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False)
+    check_bad_bbox(data_voc2, test_op, InvalidBBoxType.WrongShape, "4 features")
+
+
+if __name__ == "__main__":
+    test_random_resized_crop_with_bbox_op_c(plot_vis=True)
+    test_random_resized_crop_with_bbox_op_coco_c(plot_vis=True)
+    test_random_resized_crop_with_bbox_op_edge_c(plot_vis=True)
+    test_random_resized_crop_with_bbox_op_invalid_c()
+    test_random_resized_crop_with_bbox_op_invalid2_c()
+    test_random_resized_crop_with_bbox_op_bad_c()

tests/ut/python/dataset/test_random_crop_with_bbox.py

+# 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.
+# ==============================================================================
+"""
+Testing RandomCropWithBBox op in DE
+"""
+import numpy as np
+import mindspore.dataset as ds
+import mindspore.dataset.transforms.vision.c_transforms as c_vision
+import mindspore.dataset.transforms.vision.utils as mode
+
+from mindspore import log as logger
+from util import visualize_with_bounding_boxes, InvalidBBoxType, check_bad_bbox, \
+    config_get_set_seed, config_get_set_num_parallel_workers, save_and_check_md5
+
+GENERATE_GOLDEN = False
+
+# Updated VOC dataset with correct annotations - DATA_DIR
+DATA_DIR_VOC = "../data/dataset/testVOC2012_2"
+# COCO dataset - DATA_DIR, ANNOTATION_DIR
+DATA_DIR_COCO = ["../data/dataset/testCOCO/train/", "../data/dataset/testCOCO/annotations/train.json"]
+
+
+def test_random_crop_with_bbox_op_c(plot_vis=False):
+    """
+    Prints images and bboxes side by side with and without RandomCropWithBBox Op applied
+    """
+    logger.info("test_random_crop_with_bbox_op_c")
+
+    # Load dataset
+    dataVoc1 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False)
+    dataVoc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False)
+
+    # define test OP with values to match existing Op UT
+    test_op = c_vision.RandomCropWithBBox([512, 512], [200, 200, 200, 200])
+
+    # map to apply ops
+    dataVoc2 = dataVoc2.map(input_columns=["image", "annotation"],
+                            output_columns=["image", "annotation"],
+                            columns_order=["image", "annotation"],
+                            operations=[test_op])  # Add column for "annotation"
+
+    unaugSamp, augSamp = [], []
+
+    for unAug, Aug in zip(dataVoc1.create_dict_iterator(), dataVoc2.create_dict_iterator()):
+        unaugSamp.append(unAug)
+        augSamp.append(Aug)
+
+    if plot_vis:
+        visualize_with_bounding_boxes(unaugSamp, augSamp)
+
+
+def test_random_crop_with_bbox_op_coco_c(plot_vis=False):
+    """
+    Prints images and bboxes side by side with and without RandomCropWithBBox Op applied,
+    Testing with Coco dataset
+    """
+    logger.info("test_random_crop_with_bbox_op_coco_c")
+    # load dataset
+    dataCoco1 = ds.CocoDataset(DATA_DIR_COCO[0], annotation_file=DATA_DIR_COCO[1], task="Detection",
+                               decode=True, shuffle=False)
+
+    dataCoco2 = ds.CocoDataset(DATA_DIR_COCO[0], annotation_file=DATA_DIR_COCO[1], task="Detection",
+                               decode=True, shuffle=False)
+
+    test_op = c_vision.RandomCropWithBBox([512, 512], [200, 200, 200, 200])
+
+    dataCoco2 = dataCoco2.map(input_columns=["image", "bbox"],
+                              output_columns=["image", "bbox"],
+                              columns_order=["image", "bbox"],
+                              operations=[test_op])
+
+    unaugSamp, augSamp = [], []
+
+    for unAug, Aug in zip(dataCoco1.create_dict_iterator(), dataCoco2.create_dict_iterator()):
+        unaugSamp.append(unAug)
+        augSamp.append(Aug)
+
+    if plot_vis:
+        visualize_with_bounding_boxes(unaugSamp, augSamp, "bbox")
+
+
+def test_random_crop_with_bbox_op2_c(plot_vis=False):
+    """
+    Prints images and bboxes side by side with and without RandomCropWithBBox Op applied,
+    with md5 check, expected to pass
+    """
+    logger.info("test_random_crop_with_bbox_op2_c")
+    original_seed = config_get_set_seed(593447)
+    original_num_parallel_workers = config_get_set_num_parallel_workers(1)
+
+    # Load dataset
+    dataVoc1 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False)
+    dataVoc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False)
+
+    # define test OP with values to match existing Op unit - test
+    test_op = c_vision.RandomCropWithBBox(512, [200, 200, 200, 200], fill_value=(255, 255, 255))
+
+    # map to apply ops
+    dataVoc2 = dataVoc2.map(input_columns=["image", "annotation"],
+                            output_columns=["image", "annotation"],
+                            columns_order=["image", "annotation"],
+                            operations=[test_op])
+
+    filename = "random_crop_with_bbox_01_c_result.npz"
+    save_and_check_md5(dataVoc2, filename, generate_golden=GENERATE_GOLDEN)
+
+    unaugSamp, augSamp = [], []
+
+    for unAug, Aug in zip(dataVoc1.create_dict_iterator(), dataVoc2.create_dict_iterator()):
+        unaugSamp.append(unAug)
+        augSamp.append(Aug)
+
+    if plot_vis:
+        visualize_with_bounding_boxes(unaugSamp, augSamp)
+
+    # Restore config setting
+    ds.config.set_seed(original_seed)
+    ds.config.set_num_parallel_workers(original_num_parallel_workers)
+
+
+def test_random_crop_with_bbox_op3_c(plot_vis=False):
+    """
+    Prints images and bboxes side by side with and without RandomCropWithBBox Op applied,
+    with Padding Mode explicitly passed
+    """
+    logger.info("test_random_crop_with_bbox_op3_c")
+
+    # Load dataset
+    dataVoc1 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False)
+    dataVoc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False)
+
+    # define test OP with values to match existing Op unit - test
+    test_op = c_vision.RandomCropWithBBox(512, [200, 200, 200, 200], padding_mode=mode.Border.EDGE)
+
+    # map to apply ops
+    dataVoc2 = dataVoc2.map(input_columns=["image", "annotation"],
+                            output_columns=["image", "annotation"],
+                            columns_order=["image", "annotation"],
+                            operations=[test_op])
+
+    unaugSamp, augSamp = [], []
+
+    for unAug, Aug in zip(dataVoc1.create_dict_iterator(), dataVoc2.create_dict_iterator()):
+        unaugSamp.append(unAug)
+        augSamp.append(Aug)
+
+    if plot_vis:
+        visualize_with_bounding_boxes(unaugSamp, augSamp)
+
+
+def test_random_crop_with_bbox_op_edge_c(plot_vis=False):
+    """
+    Prints images and bboxes side by side with and without RandomCropWithBBox Op applied,
+    applied on dynamically generated edge case, expected to pass
+    """
+    logger.info("test_random_crop_with_bbox_op_edge_c")
+
+    # Load dataset
+    dataVoc1 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False)
+    dataVoc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False)
+
+    # define test OP with values to match existing Op unit - test
+    test_op = c_vision.RandomCropWithBBox(512, [200, 200, 200, 200], padding_mode=mode.Border.EDGE)
+
+    # maps to convert data into valid edge case data
+    dataVoc1 = dataVoc1.map(input_columns=["image", "annotation"],
+                            output_columns=["image", "annotation"],
+                            columns_order=["image", "annotation"],
+                            operations=[lambda img, bboxes: (img, np.array([[0, 0, img.shape[1], img.shape[0]]]).astype(bboxes.dtype))])
+
+    # Test Op added to list of Operations here
+    dataVoc2 = dataVoc2.map(input_columns=["image", "annotation"],
+                            output_columns=["image", "annotation"],
+                            columns_order=["image", "annotation"],
+                            operations=[lambda img, bboxes: (img, np.array([[0, 0, img.shape[1], img.shape[0]]]).astype(bboxes.dtype)), test_op])
+
+    unaugSamp, augSamp = [], []
+
+    for unAug, Aug in zip(dataVoc1.create_dict_iterator(), dataVoc2.create_dict_iterator()):
+        unaugSamp.append(unAug)
+        augSamp.append(Aug)
+
+    if plot_vis:
+        visualize_with_bounding_boxes(unaugSamp, augSamp)
+
+
+def test_random_crop_with_bbox_op_invalid_c():
+    """
+    Test RandomCropWithBBox Op on invalid constructor parameters, expected to raise ValueError
+    """
+    logger.info("test_random_crop_with_bbox_op_invalid_c")
+
+    # Load dataset
+    dataVoc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False)
+
+    try:
+        # define test OP with values to match existing Op unit - test
+        test_op = c_vision.RandomCropWithBBox([512, 512, 375])
+
+        # map to apply ops
+        dataVoc2 = dataVoc2.map(input_columns=["image", "annotation"],
+                                output_columns=["image", "annotation"],
+                                columns_order=["image", "annotation"],
+                                operations=[test_op])  # Add column for "annotation"
+
+        for _ in dataVoc2.create_dict_iterator():
+            break
+    except TypeError as err:
+        logger.info("Got an exception in DE: {}".format(str(err)))
+        assert "Size should be a single integer" in str(err)
+
+
+def test_random_crop_with_bbox_op_bad_c():
+    """
+    Tests RandomCropWithBBox Op with invalid bounding boxes, expected to catch multiple errors.
+    """
+    logger.info("test_random_crop_with_bbox_op_bad_c")
+    test_op = c_vision.RandomCropWithBBox([512, 512], [200, 200, 200, 200])
+
+    data_voc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False)
+    check_bad_bbox(data_voc2, test_op, InvalidBBoxType.WidthOverflow, "bounding boxes is out of bounds of the image")
+    data_voc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False)
+    check_bad_bbox(data_voc2, test_op, InvalidBBoxType.HeightOverflow, "bounding boxes is out of bounds of the image")
+    data_voc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False)
+    check_bad_bbox(data_voc2, test_op, InvalidBBoxType.NegativeXY, "min_x")
+    data_voc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False)
+    check_bad_bbox(data_voc2, test_op, InvalidBBoxType.WrongShape, "4 features")
+
+
+if __name__ == "__main__":
+    test_random_crop_with_bbox_op_c(plot_vis=True)
+    test_random_crop_with_bbox_op_coco_c(plot_vis=True)
+    test_random_crop_with_bbox_op2_c(plot_vis=True)
+    test_random_crop_with_bbox_op3_c(plot_vis=True)
+    test_random_crop_with_bbox_op_edge_c(plot_vis=True)
+    test_random_crop_with_bbox_op_invalid_c()
+    test_random_crop_with_bbox_op_bad_c()

tests/ut/python/dataset/test_random_vertical_flip_with_bbox.py

+# 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.
+# ==============================================================================
+"""
+Testing RandomVerticalFlipWithBBox op in DE
+"""
+import numpy as np
+import mindspore.dataset as ds
+import mindspore.dataset.transforms.vision.c_transforms as c_vision
+
+from mindspore import log as logger
+from util import visualize_with_bounding_boxes, InvalidBBoxType, check_bad_bbox, \
+    config_get_set_seed, config_get_set_num_parallel_workers, save_and_check_md5
+
+GENERATE_GOLDEN = False
+
+# Updated VOC dataset with correct annotations - DATA_DIR
+DATA_DIR_VOC = "../data/dataset/testVOC2012_2"
+# COCO dataset - DATA_DIR, ANNOTATION_DIR
+DATA_DIR_COCO = ["../data/dataset/testCOCO/train/", "../data/dataset/testCOCO/annotations/train.json"]
+
+
+def test_random_vertical_flip_with_bbox_op_c(plot_vis=False):
+    """
+    Prints images and bboxes side by side with and without RandomVerticalFlipWithBBox Op applied
+    """
+    logger.info("test_random_vertical_flip_with_bbox_op_c")
+    # Load dataset
+    dataVoc1 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train",
+                             decode=True, shuffle=False)
+
+    dataVoc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train",
+                             decode=True, shuffle=False)
+
+    test_op = c_vision.RandomVerticalFlipWithBBox(1)
+
+    # map to apply ops
+    dataVoc2 = dataVoc2.map(input_columns=["image", "annotation"],
+                            output_columns=["image", "annotation"],
+                            columns_order=["image", "annotation"],
+                            operations=[test_op])
+
+    unaugSamp, augSamp = [], []
+
+    for unAug, Aug in zip(dataVoc1.create_dict_iterator(), dataVoc2.create_dict_iterator()):
+        unaugSamp.append(unAug)
+        augSamp.append(Aug)
+
+    if plot_vis:
+        visualize_with_bounding_boxes(unaugSamp, augSamp)
+
+def test_random_vertical_flip_with_bbox_op_coco_c(plot_vis=False):
+    """
+    Prints images and bboxes side by side with and without RandomVerticalFlipWithBBox Op applied,
+    Testing with Coco dataset
+    """
+    logger.info("test_random_vertical_flip_with_bbox_op_coco_c")
+    # load dataset
+    dataCoco1 = ds.CocoDataset(DATA_DIR_COCO[0], annotation_file=DATA_DIR_COCO[1], task="Detection",
+                               decode=True, shuffle=False)
+
+    dataCoco2 = ds.CocoDataset(DATA_DIR_COCO[0], annotation_file=DATA_DIR_COCO[1], task="Detection",
+                               decode=True, shuffle=False)
+
+    test_op = c_vision.RandomVerticalFlipWithBBox(1)
+
+    dataCoco2 = dataCoco2.map(input_columns=["image", "bbox"],
+                              output_columns=["image", "bbox"],
+                              columns_order=["image", "bbox"],
+                              operations=[test_op])
+
+    test_op = c_vision.RandomVerticalFlipWithBBox(1)
+
+    unaugSamp, augSamp = [], []
+
+    for unAug, Aug in zip(dataCoco1.create_dict_iterator(), dataCoco2.create_dict_iterator()):
+        unaugSamp.append(unAug)
+        augSamp.append(Aug)
+
+    if plot_vis:
+        visualize_with_bounding_boxes(unaugSamp, augSamp, "bbox")
+
+
+def test_random_vertical_flip_with_bbox_op_rand_c(plot_vis=False):
+    """
+    Prints images and bboxes side by side with and without RandomVerticalFlipWithBBox Op applied,
+    tests with MD5 check, expected to pass
+    """
+    logger.info("test_random_vertical_flip_with_bbox_op_rand_c")
+    original_seed = config_get_set_seed(29847)
+    original_num_parallel_workers = config_get_set_num_parallel_workers(1)
+
+    # Load dataset
+    dataVoc1 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train",
+                             decode=True, shuffle=False)
+
+    dataVoc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train",
+                             decode=True, shuffle=False)
+
+    test_op = c_vision.RandomVerticalFlipWithBBox(0.8)
+
+    # map to apply ops
+    dataVoc2 = dataVoc2.map(input_columns=["image", "annotation"],
+                            output_columns=["image", "annotation"],
+                            columns_order=["image", "annotation"],
+                            operations=[test_op])
+
+    filename = "random_vertical_flip_with_bbox_01_c_result.npz"
+    save_and_check_md5(dataVoc2, filename, generate_golden=GENERATE_GOLDEN)
+
+    unaugSamp, augSamp = [], []
+
+    for unAug, Aug in zip(dataVoc1.create_dict_iterator(), dataVoc2.create_dict_iterator()):
+        unaugSamp.append(unAug)
+        augSamp.append(Aug)
+
+    if plot_vis:
+        visualize_with_bounding_boxes(unaugSamp, augSamp)
+
+    # Restore config setting
+    ds.config.set_seed(original_seed)
+    ds.config.set_num_parallel_workers(original_num_parallel_workers)
+
+
+def test_random_vertical_flip_with_bbox_op_edge_c(plot_vis=False):
+    """
+    Prints images and bboxes side by side with and without RandomVerticalFlipWithBBox Op applied,
+    applied on dynamically generated edge case, expected to pass
+    """
+    logger.info("test_random_vertical_flip_with_bbox_op_edge_c")
+    dataVoc1 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train",
+                             decode=True, shuffle=False)
+
+    dataVoc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train",
+                             decode=True, shuffle=False)
+
+    test_op = c_vision.RandomVerticalFlipWithBBox(1)
+
+    # maps to convert data into valid edge case data
+    dataVoc1 = dataVoc1.map(input_columns=["image", "annotation"],
+                            output_columns=["image", "annotation"],
+                            columns_order=["image", "annotation"],
+                            operations=[lambda img, bboxes: (img, np.array([[0, 0, img.shape[1], img.shape[0]]]).astype(bboxes.dtype))])
+
+    # Test Op added to list of Operations here
+    dataVoc2 = dataVoc2.map(input_columns=["image", "annotation"],
+                            output_columns=["image", "annotation"],
+                            columns_order=["image", "annotation"],
+                            operations=[lambda img, bboxes: (img, np.array([[0, 0, img.shape[1], img.shape[0]]]).astype(bboxes.dtype)), test_op])
+
+    unaugSamp, augSamp = [], []
+
+    for unAug, Aug in zip(dataVoc1.create_dict_iterator(), dataVoc2.create_dict_iterator()):
+        unaugSamp.append(unAug)
+        augSamp.append(Aug)
+
+    if plot_vis:
+        visualize_with_bounding_boxes(unaugSamp, augSamp)
+
+
+def test_random_vertical_flip_with_bbox_op_invalid_c():
+    """
+    Test RandomVerticalFlipWithBBox Op on invalid constructor parameters, expected to raise ValueError
+    """
+    logger.info("test_random_vertical_flip_with_bbox_op_invalid_c")
+    dataVoc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train",
+                             decode=True, shuffle=False)
+
+    try:
+        test_op = c_vision.RandomVerticalFlipWithBBox(2)
+
+        # map to apply ops
+        dataVoc2 = dataVoc2.map(input_columns=["image", "annotation"],
+                                output_columns=["image", "annotation"],
+                                columns_order=["image", "annotation"],
+                                operations=[test_op])
+
+        for _ in dataVoc2.create_dict_iterator():
+            break
+
+    except ValueError as err:
+        logger.info("Got an exception in DE: {}".format(str(err)))
+        assert "Input is not" in str(err)
+
+
+def test_random_vertical_flip_with_bbox_op_bad_c():
+    """
+    Tests RandomVerticalFlipWithBBox Op with invalid bounding boxes, expected to catch multiple errors
+    """
+    logger.info("test_random_vertical_flip_with_bbox_op_bad_c")
+    test_op = c_vision.RandomVerticalFlipWithBBox(1)
+
+    data_voc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False)
+    check_bad_bbox(data_voc2, test_op, InvalidBBoxType.WidthOverflow, "bounding boxes is out of bounds of the image")
+    data_voc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False)
+    check_bad_bbox(data_voc2, test_op, InvalidBBoxType.HeightOverflow, "bounding boxes is out of bounds of the image")
+    data_voc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False)
+    check_bad_bbox(data_voc2, test_op, InvalidBBoxType.NegativeXY, "min_x")
+    data_voc2 = ds.VOCDataset(DATA_DIR_VOC, task="Detection", mode="train", decode=True, shuffle=False)
+    check_bad_bbox(data_voc2, test_op, InvalidBBoxType.WrongShape, "4 features")
+
+
+if __name__ == "__main__":
+    test_random_vertical_flip_with_bbox_op_c(plot_vis=True)
+    test_random_vertical_flip_with_bbox_op_coco_c(plot_vis=True)
+    test_random_vertical_flip_with_bbox_op_rand_c(plot_vis=True)
+    test_random_vertical_flip_with_bbox_op_edge_c(plot_vis=True)
+    test_random_vertical_flip_with_bbox_op_invalid_c()
+    test_random_vertical_flip_with_bbox_op_bad_c()

tests/ut/python/dataset/util.py

@@ -288,12 +288,13 @@ def config_get_set_num_parallel_workers(num_parallel_workers_new):
     return num_parallel_workers_original
 
 
-def visualize_with_bounding_boxes(orig, aug, plot_rows=3):
+def visualize_with_bounding_boxes(orig, aug, annot_name="annotation", plot_rows=3):
     """
     Take a list of un-augmented and augmented images with "annotation" bounding boxes
     Plot images to compare test correct BBox augment functionality
     :param orig: list of original images and bboxes (without aug)
     :param aug: list of augmented images and bboxes
+    :param annot_name: the dict key for bboxes in data, e.g "bbox" (COCO) / "annotation" (VOC)
     :param plot_rows: number of rows on plot (rows = samples on one plot)
     :return: None
     """
@@ -301,9 +302,10 @@ def visualize_with_bounding_boxes(orig, aug, plot_rows=3):
     def add_bounding_boxes(ax, bboxes):
         for bbox in bboxes:
             rect = patches.Rectangle((bbox[0], bbox[1]),
-                                     bbox[2], bbox[3],
-                                     linewidth=1, edgecolor='r', facecolor='none')
+                                     bbox[2]*0.997, bbox[3]*0.997,
+                                     linewidth=1.80, edgecolor='r', facecolor='none')
             # Add the patch to the Axes
+            # Params to Rectangle slightly modified to prevent drawing overflow
             ax.add_patch(rect)
 
     # Quick check to confirm correct input parameters
@@ -337,15 +339,15 @@ def visualize_with_bounding_boxes(orig, aug, plot_rows=3):
             (axA, axB) = (axs[x, 0], axs[x, 1]) if (curPlot > 1) else (axs[0], axs[1])  # select plotting axes based on number of image rows on plot - else case when 1 row
 
             axA.imshow(dataA["image"])
-            add_bounding_boxes(axA, dataA["annotation"])
+            add_bounding_boxes(axA, dataA[annot_name])
             axA.title.set_text("Original" + str(cur_ix+1))
 
             axB.imshow(dataB["image"])
-            add_bounding_boxes(axB, dataB["annotation"])
+            add_bounding_boxes(axB, dataB[annot_name])
             axB.title.set_text("Augmented" + str(cur_ix+1))
 
-            logger.info("Original **\n{} : {}".format(str(cur_ix+1), dataA["annotation"]))
-            logger.info("Augmented **\n{} : {}\n".format(str(cur_ix+1), dataB["annotation"]))
+            logger.info("Original **\n{} : {}".format(str(cur_ix+1), dataA[annot_name]))
+            logger.info("Augmented **\n{} : {}\n".format(str(cur_ix+1), dataB[annot_name]))
 
         plt.show()
 
@@ -381,19 +383,19 @@ def check_bad_bbox(data, test_op, invalid_bbox_type, expected_error):
         width = img.shape[1]
         if invalid_bbox_type_ == InvalidBBoxType.WidthOverflow:
             # use box that overflows on width
-            return img, np.array([[0, 0, width + 1, height, 0, 0, 0]]).astype(np.uint32)
+            return img, np.array([[0, 0, width + 1, height, 0, 0, 0]]).astype(np.float32)
 
         if invalid_bbox_type_ == InvalidBBoxType.HeightOverflow:
             # use box that overflows on height
-            return img, np.array([[0, 0, width, height + 1, 0, 0, 0]]).astype(np.uint32)
+            return img, np.array([[0, 0, width, height + 1, 0, 0, 0]]).astype(np.float32)
 
         if invalid_bbox_type_ == InvalidBBoxType.NegativeXY:
             # use box with negative xy
-            return img, np.array([[-10, -10, width, height, 0, 0, 0]]).astype(np.uint32)
+            return img, np.array([[-10, -10, width, height, 0, 0, 0]]).astype(np.float32)
 
         if invalid_bbox_type_ == InvalidBBoxType.WrongShape:
             # use box that has incorrect shape
-            return img, np.array([[0, 0, width - 1]]).astype(np.uint32)
+            return img, np.array([[0, 0, width - 1]]).astype(np.float32)
         return img, bboxes
 
     try: