def get_vis(vis, _gt_boxes, _prediction_dict):
vis_gt_boxes = np.zeros((len(_gt_boxes), 100, 4))
for ib in range(len(_gt_boxes)):
for ij in range(len(_gt_boxes[ib])):
vis_gt_boxes[ib, ij, :] = _gt_boxes[ib][ij]
#print('vis_gt_scores',vis_gt_scores)
vis_gt_classes = np.zeros((len(_gt_boxes), 100), 'uint32')
for ib in range(len(_gt_boxes)):
for ij in range(len(_gt_boxes[ib])):
vis_gt_classes[ib][ij] = 2
vis_gt_scores = np.zeros((len(_gt_boxes), 100))
for ib in range(len(_gt_boxes)):
for ij in range(len(_gt_boxes[ib])):
vis_gt_scores[ib][ij] = 1.
vis = viz_utils.draw_bounding_boxes_on_image_tensors(
vis,
vis_gt_boxes, # (4, 100, 4) detection_classes (4, 100) (4, 100)
vis_gt_classes,
vis_gt_scores, #np.ones((4,len(gt_boxes), 100)),
category_index)
vis = tf.cast(vis, tf.uint8)
vis = viz_utils.draw_bounding_boxes_on_image_tensors(
vis, #tf.cast(tf.concat(image_tensors,axis=0),tf.uint8), #vis,
_prediction_dict['detection_boxes'],
_prediction_dict['detection_classes'].numpy().astype(
np.uint32) +
label_id_offset, #prediction_dict['detection_classes'],#prediction_dict['detection_classes'].astype(tf.int32) + label_id_offset,
_prediction_dict['detection_scores'],
category_index)
vis = tf.cast(vis, tf.float32)
return vis
def test_draw_bounding_boxes_on_image_tensors(self):
"""Tests that bounding box utility produces reasonable results."""
category_index = {1: {'id': 1, 'name': 'dog'}, 2: {'id': 2, 'name': 'cat'}}
fname = os.path.join(_TESTDATA_PATH, 'image1.jpg')
image_np = np.array(Image.open(fname))
images_np = np.stack((image_np, image_np), axis=0)
with tf.Graph().as_default():
images_tensor = tf.constant(value=images_np, dtype=tf.uint8)
boxes = tf.constant([[[0.4, 0.25, 0.75, 0.75], [0.5, 0.3, 0.6, 0.9]],
[[0.25, 0.25, 0.75, 0.75], [0.1, 0.3, 0.6, 1.0]]])
classes = tf.constant([[1, 1], [1, 2]], dtype=tf.int64)
scores = tf.constant([[0.8, 0.1], [0.6, 0.5]])
images_with_boxes = (
visualization_utils.draw_bounding_boxes_on_image_tensors(
images_tensor,
boxes,
classes,
scores,
category_index,
min_score_thresh=0.2))
with self.test_session() as sess:
sess.run(tf.global_variables_initializer())
# Write output images for visualization.
images_with_boxes_np = sess.run(images_with_boxes)
self.assertEqual(images_np.shape, images_with_boxes_np.shape)
for i in range(images_with_boxes_np.shape[0]):
img_name = 'image_' + str(i) + '.png'
output_file = os.path.join(self.get_temp_dir(), img_name)
logging.info('Writing output image %d to %s', i, output_file)
image_pil = Image.fromarray(images_with_boxes_np[i, ...])
image_pil.save(output_file)
def graph_fn():
images_tensor = tf.constant(value=images_np, dtype=tf.uint8)
image_shape = tf.constant(original_image_shape, dtype=tf.int32)
boxes = tf.constant([[[0.4, 0.25, 0.75, 0.75],
[0.5, 0.3, 0.6, 0.9]],
[[0.25, 0.25, 0.75, 0.75],
[0.1, 0.3, 0.6, 1.0]]])
classes = tf.constant([[1, 1], [1, 2]], dtype=tf.int64)
scores = tf.constant([[0.8, 0.1], [0.6, 0.5]])
keypoints = tf.random.uniform((2, 2, 4, 2),
maxval=1.0,
dtype=tf.float32)
keypoint_edges = [(0, 1), (1, 2), (2, 3), (3, 0)]
images_with_boxes = (
visualization_utils.draw_bounding_boxes_on_image_tensors(
images_tensor,
boxes,
classes,
scores,
category_index,
original_image_spatial_shape=image_shape,
true_image_shape=image_shape,
keypoints=keypoints,
min_score_thresh=0.2,
keypoint_edges=keypoint_edges))
return images_with_boxes
def test_draw_bounding_boxes_on_image_tensors_grayscale(self):
"""Tests the case where input image tensor has one channel."""
category_index = {1: {'id': 1, 'name': 'dog'}}
image_np = self.create_test_grayscale_image()
images_np = np.stack((image_np, image_np), axis=0)
with tf.Graph().as_default():
images_tensor = tf.constant(value=images_np, dtype=tf.uint8)
image_shape = tf.constant([[100, 200], [100, 200]], dtype=tf.int32)
boxes = tf.constant(0, dtype=tf.float32, shape=[2, 0, 4])
classes = tf.constant(0, dtype=tf.int64, shape=[2, 0])
scores = tf.constant(0, dtype=tf.float32, shape=[2, 0])
images_with_boxes = (
visualization_utils.draw_bounding_boxes_on_image_tensors(
images_tensor,
boxes,
classes,
scores,
category_index,
original_image_spatial_shape=image_shape,
true_image_shape=image_shape,
min_score_thresh=0.2))
with self.test_session() as sess:
sess.run(tf.global_variables_initializer())
final_images_np = sess.run(images_with_boxes)
self.assertEqual((2, 100, 200, 3), final_images_np.shape)
def test_draw_bounding_boxes_on_image_tensors_with_additional_channels(self):
"""Tests the case where input image tensor has more than 3 channels."""
category_index = {1: {'id': 1, 'name': 'dog'}}
image_np = self.create_test_image_with_five_channels()
images_np = np.stack((image_np, image_np), axis=0)
with tf.Graph().as_default():
images_tensor = tf.constant(value=images_np, dtype=tf.uint8)
boxes = tf.constant(0, dtype=tf.float32, shape=[2, 0, 4])
classes = tf.constant(0, dtype=tf.int64, shape=[2, 0])
scores = tf.constant(0, dtype=tf.float32, shape=[2, 0])
images_with_boxes = (
visualization_utils.draw_bounding_boxes_on_image_tensors(
images_tensor,
boxes,
classes,
scores,
category_index,
min_score_thresh=0.2))
with self.test_session() as sess:
sess.run(tf.global_variables_initializer())
final_images_np = sess.run(images_with_boxes)
self.assertEqual((2, 100, 200, 3), final_images_np.shape)
def test_draw_bounding_boxes_on_image_tensors(self):
"""Tests that bounding box utility produces reasonable results."""
category_index = {1: {'id': 1, 'name': 'dog'}, 2: {'id': 2, 'name': 'cat'}}
fname = os.path.join(_TESTDATA_PATH, 'image1.jpg')
image_np = np.array(Image.open(fname))
images_np = np.stack((image_np, image_np), axis=0)
with tf.Graph().as_default():
images_tensor = tf.constant(value=images_np, dtype=tf.uint8)
boxes = tf.constant([[[0.4, 0.25, 0.75, 0.75], [0.5, 0.3, 0.6, 0.9]],
[[0.25, 0.25, 0.75, 0.75], [0.1, 0.3, 0.6, 1.0]]])
classes = tf.constant([[1, 1], [1, 2]], dtype=tf.int64)
scores = tf.constant([[0.8, 0.1], [0.6, 0.5]])
images_with_boxes = (
visualization_utils.draw_bounding_boxes_on_image_tensors(
images_tensor,
boxes,
classes,
scores,
category_index,
min_score_thresh=0.2))
with self.test_session() as sess:
sess.run(tf.global_variables_initializer())
# Write output images for visualization.
images_with_boxes_np = sess.run(images_with_boxes)
self.assertEqual(images_np.shape, images_with_boxes_np.shape)
for i in range(images_with_boxes_np.shape[0]):
img_name = 'image_' + str(i) + '.png'
output_file = os.path.join(self.get_temp_dir(), img_name)
logging.info('Writing output image %d to %s', i, output_file)
image_pil = Image.fromarray(images_with_boxes_np[i, ...])
image_pil.save(output_file)
def graph_fn():
images_tensor = tf.constant(value=images_np, dtype=tf.uint8)
boxes = tf.constant(0, dtype=tf.float32, shape=[2, 0, 4])
classes = tf.constant(0, dtype=tf.int64, shape=[2, 0])
scores = tf.constant(0, dtype=tf.float32, shape=[2, 0])
images_with_boxes = (
visualization_utils.draw_bounding_boxes_on_image_tensors(
images_tensor,
boxes,
classes,
scores,
category_index,
min_score_thresh=0.2))
return images_with_boxes
def process(
self,
images,
object_boxes,
object_scores=None,
object_classes=None,
object_instance_ids=None,
object_keypoints=None,
object_instance_masks_on_image=None,
):
# pylint: disable=arguments-differ,too-many-arguments
# base class has more generic signature
images = tf.cast(images * 256.0, tf.uint8)
if object_scores is None:
object_scores = tf.ones_like(object_boxes, tf.float32)[..., 0]
if object_classes is None:
ones = tf.ones_like(object_scores, tf.int64)
zeros = tf.zeros_like(object_scores, tf.int64)
object_classes = tf.where(object_scores > 0, ones, zeros)
object_classes = tf.cast(object_classes, tf.int64)
if len(images.get_shape()) == 3:
images = tf.expand_dims(images, -1)
if images.get_shape().as_list()[-1] == 1:
images = tf.tile(images, [1, 1, 1, 3])
true_image_shapes = get_true_image_shapes(images)
image_with_detections = draw_bounding_boxes_on_image_tensors(
images=images,
boxes=object_boxes,
classes=object_classes,
scores=object_scores,
keypoints=object_keypoints,
instance_masks=object_instance_masks_on_image,
category_index=self.category_index,
track_ids=object_instance_ids,
max_boxes_to_draw=self.max_boxes_to_draw,
true_image_shape=true_image_shapes,
original_image_spatial_shape=true_image_shapes[:, :2],
min_score_thresh=self.score_threshold,
use_normalized_coordinates=self.use_normalized_coordinates,
)
image_with_detections = tf.reshape(image_with_detections,
tf.shape(images))
result = {ObjectDataFields.images_with_objects: image_with_detections}
return result
def graph_fn():
images_tensor = tf.constant(value=images_np, dtype=tf.uint8)
image_shape = tf.constant(original_image_shape, dtype=tf.int32)
boxes = tf.constant([[[0.4, 0.25, 0.75, 0.75],
[0.5, 0.3, 0.7, 0.9], [0.7, 0.5, 0.8, 0.9]],
[[0.41, 0.25, 0.75, 0.75],
[0.51, 0.3, 0.7, 0.9], [0.75, 0.5, 0.8,
0.9]]])
classes = tf.constant([[1, 1, 2], [1, 1, 2]], dtype=tf.int64)
scores = tf.constant([[0.8, 0.5, 0.7], [0.6, 0.5, 0.8]])
track_ids = tf.constant([[3, 9, 7], [3, 9, 144]], dtype=tf.int32)
images_with_boxes = (
visualization_utils.draw_bounding_boxes_on_image_tensors(
images_tensor,
boxes,
classes,
scores,
category_index,
original_image_spatial_shape=image_shape,
true_image_shape=image_shape,
track_ids=track_ids,
min_score_thresh=0.2))
return images_with_boxes
