def box_iou(self, true_box, pred_box):
# based on the type of loss, compute the iou loss for a box
# compute_<name> indicated the type of iou to use
if self.iou_type == 'giou':
_, iou = box_ops.compute_giou(true_box, pred_box)
elif self.iou_type == 'ciou':
_, iou = box_ops.compute_ciou(true_box, pred_box)
else:
iou = box_ops.compute_iou(true_box, pred_box)
return iou
def box_loss(self, true_box, pred_box, darknet=False):
"""Call iou function and use it to compute the loss for the box maps."""
if self._loss_type == 'giou':
iou, liou = box_ops.compute_giou(true_box, pred_box)
elif self._loss_type == 'ciou':
iou, liou = box_ops.compute_ciou(true_box, pred_box, darknet=darknet)
else:
liou = iou = box_ops.compute_iou(true_box, pred_box)
loss_box = 1 - liou
return iou, liou, loss_box
def test_ious(self, num_boxes):
boxes = tf.convert_to_tensor(np.random.rand(num_boxes, 4))
expected_shape = np.array([
num_boxes,
])
expected_iou = np.ones([
num_boxes,
])
iou = box_ops.compute_iou(boxes, boxes)
_, giou = box_ops.compute_giou(boxes, boxes)
_, ciou = box_ops.compute_ciou(boxes, boxes)
_, diou = box_ops.compute_diou(boxes, boxes)
self.assertAllEqual(tf.shape(iou).numpy(), expected_shape)
self.assertArrayNear(iou, expected_iou, 0.001)
self.assertArrayNear(giou, expected_iou, 0.001)
self.assertArrayNear(ciou, expected_iou, 0.001)
self.assertArrayNear(diou, expected_iou, 0.001)
def aggregated_comparative_iou(boxes1, boxes2=None, iou_type=0):
k = tf.shape(boxes1)[-2]
boxes1 = tf.expand_dims(boxes1, axis=-2)
boxes1 = tf.tile(boxes1, [1, 1, k, 1])
if boxes2 is not None:
boxes2 = tf.expand_dims(boxes2, axis=-2)
boxes2 = tf.tile(boxes2, [1, 1, k, 1])
boxes2 = tf.transpose(boxes2, perm=(0, 2, 1, 3))
else:
boxes2 = tf.transpose(boxes1, perm=(0, 2, 1, 3))
if iou_type == 0: # diou
_, iou = box_ops.compute_diou(boxes1, boxes2)
elif iou_type == 1: # giou
_, iou = box_ops.compute_giou(boxes1, boxes2)
else:
iou = box_ops.compute_iou(boxes1, boxes2, yxyx=True)
return iou
def get_best_anchor(y_true, anchors, width=1, height=1):
"""Gets the correct anchor that is assoiciated with each box using IOU.
Args:
y_true: tf.Tensor[] for the list of bounding boxes in the yolo format
anchors: list or tensor for the anchor boxes to be used in prediction
found via Kmeans
width: int for the image width
height: int for the image height
Returns:
tf.Tensor: y_true with the anchor associated with each ground truth
box known.
"""
with tf.name_scope('get_anchor'):
width = tf.cast(width, dtype=tf.float32)
height = tf.cast(height, dtype=tf.float32)
# split the boxes into center and width height
anchor_xy = y_true[..., 0:2]
# scale thhe boxes
anchors = tf.convert_to_tensor(anchors, dtype=tf.float32)
anchors_x = anchors[..., 0] / width
anchors_y = anchors[..., 1] / height
anchors = tf.stack([anchors_x, anchors_y], axis=-1)
k = tf.shape(anchors)[0]
# build a matrix of anchor boxes of shape [num_anchors, num_boxes, 4]
anchors = tf.transpose(anchors, perm=[1, 0])
anchor_xy = tf.tile(tf.expand_dims(anchor_xy, axis=-1),
[1, 1, tf.shape(anchors)[-1]])
anchors = tf.tile(tf.expand_dims(anchors, axis=0),
[tf.shape(anchor_xy)[0], 1, 1])
# stack the xy so, each anchor is asscoaited once with each center from
# the ground truth input
anchors = tf.concat([anchor_xy, anchors], axis=1)
anchors = tf.transpose(anchors, perm=[2, 0, 1])
# copy the gt n times so that each anchor from above can be compared to
# input ground truth to shape: [num_anchors, num_boxes, 4]
truth_comp = tf.tile(tf.expand_dims(y_true[..., 0:4], axis=-1),
[1, 1, tf.shape(anchors)[0]])
truth_comp = tf.transpose(truth_comp, perm=[2, 0, 1])
# compute intersection over union of the boxes, and take the argmax of
# comuted iou for each box. thus each box is associated with the
# largest interection over union
iou_raw = box_ops.compute_iou(truth_comp, anchors)
values, indexes = tf.math.top_k(tf.transpose(iou_raw, perm=[1, 0]),
k=tf.cast(k, dtype=tf.int32),
sorted=True)
ind_mask = tf.cast(values > 0.213, dtype=indexes.dtype)
# pad the indexs such that all values less than the thresh are -1
# add one, multiply the mask to zeros all the bad locations
# subtract 1 makeing all the bad locations 0.
iou_index = tf.concat([
tf.keras.backend.expand_dims(indexes[..., 0], axis=-1),
((indexes[..., 1:] + 1) * ind_mask[..., 1:]) - 1
],
axis=-1)
iou_index = iou_index[..., :6]
return tf.cast(iou_index, dtype=tf.float32)
def get_best_anchor(y_true, anchors, width=1, height=1):
"""Gets the correct anchor that is assoiciated with each box using IOU between
input anchors and ground truth.
Args:
y_true: tf.Tensor[] for the list of bounding boxes in the yolo format
anchors: list or tensor for the anchor boxes to be used in prediction
found via Kmeans
size: size of the image that the bounding boxes were selected at 416 is
the default for the original YOLO model
return:
tf.Tensor: y_true with the anchor associated with each ground truth box
known
"""
with tf.name_scope('get_anchor'):
width = tf.cast(width, dtype=tf.float32)
height = tf.cast(height, dtype=tf.float32)
anchor_xy = y_true[..., 0:2]
true_wh = y_true[..., 2:4]
# scale thhe boxes
anchors = tf.convert_to_tensor(anchors, dtype=tf.float32)
anchors_x = anchors[..., 0] / width
anchors_y = anchors[..., 1] / height
anchors = tf.stack([anchors_x, anchors_y], axis=-1)
# build a matrix of anchor boxes
anchors = tf.transpose(anchors, perm=[1, 0])
anchor_xy = tf.tile(tf.expand_dims(anchor_xy, axis=-1),
[1, 1, tf.shape(anchors)[-1]])
anchors = tf.tile(tf.expand_dims(anchors, axis=0),
[tf.shape(anchor_xy)[0], 1, 1])
# stack the xy so, each anchor is asscoaited once with each center from
# the ground truth input
anchors = tf.keras.layers.concatenate([anchor_xy, anchors], axis=1)
anchors = tf.transpose(anchors, perm=[2, 0, 1])
# copy the gt n times so that each anchor from above can be compared to
# input ground truth
truth_comp = tf.tile(tf.expand_dims(y_true[..., 0:4], axis=-1),
[1, 1, tf.shape(anchors)[0]])
truth_comp = tf.transpose(truth_comp, perm=[2, 0, 1])
# compute intersection over union of the boxes, and take the argmax of
# comuted iou for each box. thus each box is associated with the largest
# interection over union
iou_raw = box_ops.compute_iou(truth_comp, anchors)
gt_mask = tf.cast(iou_raw > 0.213, dtype=iou_raw.dtype)
num_k = tf.reduce_max(
tf.reduce_sum(tf.transpose(gt_mask, perm=[1, 0]), axis=1))
if num_k <= 0:
num_k = 1.0
values, indexes = tf.math.top_k(tf.transpose(iou_raw, perm=[1, 0]),
k=tf.cast(num_k, dtype=tf.int32),
sorted=True)
ind_mask = tf.cast(values > 0.213, dtype=indexes.dtype)
iou_index = tf.concat([
tf.expand_dims(indexes[..., 0], axis=-1),
((indexes[..., 1:] + 1) * ind_mask[..., 1:]) - 1
],
axis=-1)
stack = tf.zeros(
[tf.shape(iou_index)[0],
tf.cast(1, dtype=iou_index.dtype)],
dtype=iou_index.dtype) - 1
while num_k < 5:
iou_index = tf.concat([iou_index, stack], axis=-1)
num_k += 1
iou_index = iou_index[..., :5]
values = tf.concat([
tf.expand_dims(values[..., 0], axis=-1),
((values[..., 1:]) * tf.cast(ind_mask[..., 1:], dtype=tf.float32))
],
axis=-1)
return tf.cast(iou_index, dtype=tf.float32)