def test_find_rois_from_gt_boxes(self):
gt_object = {"class": "person", "bbox": np.array([0, 0, 10, 10])}
result = roi_tools.find_foreground_rois_from_ground_truth_boxes(
[gt_object], (30, 20))
assert 32 == len(result)
np.testing.assert_equal(np.array([0, 1, 9, 10]),
result['01910']['bbox'])
assert roi_tools.calculate_iou(gt_object["bbox"],
result['01910']['bbox']) > 0.7
expected_class = np.zeros(21)
expected_class[1] = 1
for roi in result.values():
np.testing.assert_equal(expected_class, roi["class"])
assert roi_tools.calculate_iou(gt_object["bbox"],
roi['bbox']) > 0.7
def print_iou_info(image, gt_bboxes):
rois_fore, rois_back = roi_tools.find_rois_complete(
image, gt_bboxes, 4, 1000)
count_background = 0
count_foreground = 0
for gt_bbox in gt_bboxes:
for roi in rois_fore:
iou = roi_tools.calculate_iou(gt_bbox["bbox"], roi)
if iou >= 0.5:
print(iou)
count_foreground = count_foreground + 1
else:
count_background = count_background + 1
print("Foreground rois: " + str(count_foreground))
print("Background rois: " + str(count_background))
def show_image_with_highest_iou_roi(image, gt_bboxes):
rois = roi_tools.find_rois_selective_search(image)
bboxes_to_display = []
for gt_bbox in gt_bboxes:
bboxes_to_display.append(gt_bbox)
max_iou = 0
max_roi = None
for roi in rois:
iou = roi_tools.calculate_iou(gt_bbox, roi)
if iou > max_iou:
max_iou = iou
max_roi = roi
print("Max IoU: " + str(max_iou))
bboxes_to_display.append(max_roi)
print(bboxes_to_display)
image_tools.show_image_with_bboxes(resized_image_in_pixels,
bboxes_to_display)
def test_calculate_iou_0_percent_horizontally(self):
bbox_1 = np.array([0, 0, 10, 10])
bbox_2 = np.array([10, 0, 10, 10])
iou = roi_tools.calculate_iou(bbox_1, bbox_2)
assert float("{0:.2f}".format(iou)) == 0.0
def test_calculate_iou_25_percent(self):
bbox_1 = np.array([0, 0, 10, 10])
bbox_2 = np.array([0, 5, 10, 10])
iou = roi_tools.calculate_iou(bbox_1, bbox_2)
assert float("{0:.2f}".format(iou)) == 0.33