using already pushed files

research/object_detection/core/box_list_ops.py

@@ -303,6 +303,7 @@ def iou(boxlist1, boxlist2, scope=None):
         tf.equal(intersections, 0.0),
         tf.zeros_like(intersections), tf.truediv(intersections, unions))
 
+
 def l1(boxlist1, boxlist2, scope=None):
   """Computes l1 loss (pairwise) between two boxlists.
 
@@ -317,20 +318,19 @@ def l1(boxlist1, boxlist2, scope=None):
   with tf.name_scope(scope, 'PairwiseL1'):
     ycenter1, xcenter1, h1, w1 = boxlist1.get_center_coordinates_and_sizes()
     ycenter2, xcenter2, h2, w2 = boxlist2.get_center_coordinates_and_sizes()
-    ycenters = tf.abs(tf.expand_dims(ycenter2, axis=0) - \
-        tf.expand_dims(tf.transpose(ycenter1), axis=1))
-    xcenters = tf.abs(tf.expand_dims(xcenter2, axis=0) - \
-        tf.expand_dims(tf.transpose(xcenter1), axis=1))
-    heights = tf.abs(tf.expand_dims(h2, axis=0) - \
-        tf.expand_dims(tf.transpose(h1), axis=1))
-    widths = tf.abs(tf.expand_dims(w2, axis=0) - \
-        tf.expand_dims(tf.transpose(w1), axis=1))
+    ycenters = tf.abs(tf.expand_dims(ycenter2, axis=0) - tf.expand_dims(
+        tf.transpose(ycenter1), axis=1))
+    xcenters = tf.abs(tf.expand_dims(xcenter2, axis=0) - tf.expand_dims(
+        tf.transpose(xcenter1), axis=1))
+    heights = tf.abs(tf.expand_dims(h2, axis=0) - tf.expand_dims(
+        tf.transpose(h1), axis=1))
+    widths = tf.abs(tf.expand_dims(w2, axis=0) - tf.expand_dims(
+        tf.transpose(w1), axis=1))
     return ycenters + xcenters + heights + widths
 
-def giou_loss(boxlist1, boxlist2, scope=None):
-  """
-  Computes generalized IOU loss between two boxlists pairwise,
-  as described at giou.stanford.edu.
+
+def giou(boxlist1, boxlist2, scope=None):
+  """Computes pairwise generalized IOU between two boxlists.
 
   Args:
     boxlist1: BoxList holding N boxes
@@ -340,12 +340,13 @@ def giou_loss(boxlist1, boxlist2, scope=None):
   Returns:
     a tensor with shape [N, M] representing the pairwise GIoU loss.
   """
-  with tf.name_scope(scope, "PairwiseGIoU"):
-    N = boxlist1.num_boxes()
-    M = boxlist2.num_boxes()
-    boxes1 = tf.repeat(boxlist1.get(), repeats=M, axis=0)
-    boxes2 = tf.tile(boxlist2.get(), multiples=[N, 1])
-    return tf.reshape(1.0 - ops.giou(boxes1, boxes2), [N, M])
+  with tf.name_scope(scope, 'PairwiseGIoU'):
+    n = boxlist1.num_boxes()
+    m = boxlist2.num_boxes()
+    boxes1 = tf.repeat(boxlist1.get(), repeats=m, axis=0)
+    boxes2 = tf.tile(boxlist2.get(), multiples=[n, 1])
+    return tf.reshape(ops.giou(boxes1, boxes2), [n, m])
+
 
 def matched_iou(boxlist1, boxlist2, scope=None):
   """Compute intersection-over-union between corresponding boxes in boxlists.
@@ -367,6 +368,7 @@ def matched_iou(boxlist1, boxlist2, scope=None):
         tf.equal(intersections, 0.0),
         tf.zeros_like(intersections), tf.truediv(intersections, unions))
 
+
 def ioa(boxlist1, boxlist2, scope=None):
   """Computes pairwise intersection-over-area between box collections.
 

research/object_detection/core/box_list_ops_test.py

@@ -217,6 +217,7 @@ class BoxListOpsTest(test_case.TestCase):
 
   def test_iou(self):
     def graph_fn():
+      corners1 = tf.constant([[4.0, 3.0, 7.0, 5.0], [5.0, 6.0, 10.0, 7.0]])
       corners1 = tf.constant([[4.0, 3.0, 7.0, 5.0], [5.0, 6.0, 10.0, 7.0]])
       corners2 = tf.constant([[3.0, 4.0, 6.0, 8.0], [14.0, 14.0, 15.0, 15.0],
                               [0.0, 0.0, 20.0, 20.0]])
@@ -247,9 +248,9 @@ class BoxListOpsTest(test_case.TestCase):
       corners2 = tf.constant([[5.0, 7.0, 7.0, 9.0], [5.0, 11.0, 7.0, 13.0]])
       boxes1 = box_list.BoxList(corners1)
       boxes2 = box_list.BoxList(corners2)
-      giou = box_list_ops.giou_loss(boxes1, boxes2)
+      giou = box_list_ops.giou(boxes1, boxes2)
       return giou
-    exp_output = [[0.0, 4.0 / 3.0]]
+    exp_output = [[1.0, -1.0 / 3.0]]
     giou_output = self.execute(graph_fn, [])
     self.assertAllClose(giou_output, exp_output)
 

research/object_detection/core/losses.py

@@ -211,12 +211,11 @@ class WeightedIOULocalizationLoss(Loss):
     return tf.reshape(weights, [-1]) * per_anchor_iou_loss
 
 
-
 class WeightedGIOULocalizationLoss(Loss):
-  """IOU localization loss function.
+  """GIOU localization loss function.
 
-  Sums the IOU for corresponding pairs of predicted/groundtruth boxes
-  and for each pair assign a loss of 1 - IOU.  We then compute a weighted
+  Sums the GIOU loss for corresponding pairs of predicted/groundtruth boxes
+  and for each pair assign a loss of 1 - GIOU.  We then compute a weighted
   sum over all pairs which is returned as the total loss.
   """
 

research/object_detection/core/losses_test.py

@@ -197,6 +197,7 @@ class WeightedIOULocalizationLossTest(test_case.TestCase):
     loss_output = self.execute(graph_fn, [])
     self.assertAllClose(loss_output, exp_loss)
 
+
 class WeightedGIOULocalizationLossTest(test_case.TestCase):
 
   def testReturnsCorrectLoss(self):

research/object_detection/matchers/hungarian_matcher.py

@@ -15,21 +15,16 @@
 
 """Hungarian bipartite matcher implementation."""
 
-import tensorflow.compat.v1 as tf
 import numpy as np
+from scipy.optimize import linear_sum_assignment
 
+import tensorflow.compat.v1 as tf
 from object_detection.core import matcher
-from scipy.optimize import linear_sum_assignment
 
 
 class HungarianBipartiteMatcher(matcher.Matcher):
   """Wraps a Hungarian bipartite matcher into TensorFlow."""
 
-  def __init__(self):
-    """Constructs a Matcher."""
-
-    super(HungarianBipartiteMatcher, self).__init__()
-
   def _match(self, similarity_matrix, valid_rows):
     """Optimally bipartite matches a collection rows and columns.
 

research/object_detection/matchers/hungarian_matcher_tf2_test.py

-# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
+# Copyright 2020 The TensorFlow Authors. All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -100,5 +100,6 @@ class HungarianBipartiteMatcherTest(test_case.TestCase):
     self.assertAllEqual(match_results_out._match_results.numpy(),
                         expected_match_results)
 
+
 if __name__ == '__main__':
   tf.test.main()

research/object_detection/utils/ops.py

@@ -216,13 +216,13 @@ def pad_to_multiple(tensor, multiple):
     height_pad = tf.zeros([
         batch_size, padded_tensor_height - tensor_height, tensor_width,
         tensor_depth
-    ])
+    ], dtype=tensor.dtype)
     tensor = tf.concat([tensor, height_pad], 1)
   if padded_tensor_width != tensor_width:
     width_pad = tf.zeros([
         batch_size, padded_tensor_height, padded_tensor_width - tensor_width,
         tensor_depth
-    ])
+    ], dtype=tensor.dtype)
     tensor = tf.concat([tensor, width_pad], 2)
 
   return tensor
@@ -1135,10 +1135,11 @@ def decode_image(tensor_dict):
   tensor_dict[fields.InputDataFields.image].set_shape([None, None, 3])
   return tensor_dict
 
+
 def giou(boxes1, boxes2):
-  """
-  Computes generalized IOU between two tensors. Each box should be
-  represented as [ymin, xmin, ymax, xmax].
+  """Computes generalized IOU between two tensors.
+
+  Each box should be represented as [ymin, xmin, ymax, xmax].
 
   Args:
     boxes1: a tensor with shape [num_boxes, 4]
@@ -1148,44 +1149,41 @@ def giou(boxes1, boxes2):
     a tensor of shape [num_boxes] containing GIoUs
 
   """
-  def two_boxes_giou(boxes):
+  def _two_boxes_giou(boxes):
+    """Compute giou between two boxes."""
     boxes1, boxes2 = boxes
 
-    pred_ymin = tf.gather_nd(boxes1, [0])
-    pred_xmin = tf.gather_nd(boxes1, [1])
-    pred_ymax = tf.gather_nd(boxes1, [2])
-    pred_xmax = tf.gather_nd(boxes1, [3])
-
-    gt_ymin = tf.gather_nd(boxes2, [0])
-    gt_xmin = tf.gather_nd(boxes2, [1])
-    gt_ymax = tf.gather_nd(boxes2, [2])
-    gt_xmax = tf.gather_nd(boxes2, [3])
+    pred_ymin, pred_xmin, pred_ymax, pred_xmax = tf.unstack(boxes1)
+    gt_ymin, gt_xmin, gt_ymax, gt_xmax = tf.unstack(boxes2)
 
     gt_area = (gt_ymax - gt_ymin) * (gt_xmax - gt_xmin)
     pred_area = (pred_ymax - pred_ymin) * (pred_xmax - pred_xmin)
 
-    x1I = tf.maximum(pred_xmin, gt_xmin)
-    x2I = tf.minimum(pred_xmax, gt_xmax)
-    y1I = tf.maximum(pred_ymin, gt_ymin)
-    y2I = tf.minimum(pred_ymax, gt_ymax)
-    intersection_area = (y2I - y1I) * (x2I - x1I) if (y2I > y1I and
-                                                      x2I > x1I) else 0.0
+    x1_i = tf.maximum(pred_xmin, gt_xmin)
+    x2_i = tf.minimum(pred_xmax, gt_xmax)
+    y1_i = tf.maximum(pred_ymin, gt_ymin)
+    y2_i = tf.minimum(pred_ymax, gt_ymax)
+    intersection_area = tf.maximum(0.0, y2_i - y1_i) * tf.maximum(0.0,
+                                                                  x2_i - x1_i)
 
-    x1C = tf.minimum(pred_xmin, gt_xmin)
-    x2C = tf.maximum(pred_xmax, gt_xmax)
-    y1C = tf.minimum(pred_ymin, gt_ymin)
-    y2C = tf.maximum(pred_ymax, gt_ymax)
-    hull_area = (y2C - y1C) * (x2C - x1C)
+    x1_c = tf.minimum(pred_xmin, gt_xmin)
+    x2_c = tf.maximum(pred_xmax, gt_xmax)
+    y1_c = tf.minimum(pred_ymin, gt_ymin)
+    y2_c = tf.maximum(pred_ymax, gt_ymax)
+    hull_area = (y2_c - y1_c) * (x2_c - x1_c)
 
     union_area = gt_area + pred_area - intersection_area
-    IoU = intersection_area/union_area if union_area != 0.0 else 1.0
-    gIoU = IoU - (hull_area - union_area) / hull_area if hull_area != 0.0 else IoU
+    iou = tf.where(
+        tf.equal(union_area, 0.0), 0.0, intersection_area / union_area)
+    giou_ = iou - tf.where(hull_area > 0.0,
+                           (hull_area - union_area) / hull_area, iou)
+
+    return giou_
 
-    return gIoU
+  return shape_utils.static_or_dynamic_map_fn(_two_boxes_giou, [boxes1, boxes2])
 
-  return shape_utils.static_or_dynamic_map_fn(two_boxes_giou, [boxes1, boxes2])
 
-def cy_cx_h_w_to_ymin_xmin_ymax_xmax(input_tensor):
+def center_to_corner_coordinate(input_tensor):
   """Converts input boxes from center to corner representation."""
   reshaped_encodings = tf.reshape(input_tensor, [-1, 4])
   ycenter = tf.gather(reshaped_encodings, [0], axis=1)

research/object_detection/utils/ops_test.py

@@ -1630,20 +1630,95 @@ class TestGatherWithPaddingValues(test_case.TestCase):
     self.assertAllClose(expected_gathered_tensor, gathered_tensor_np)
 
 
+
+
+
+
+
+
 class TestGIoU(test_case.TestCase):
 
-  def test_giou(self):
+  def test_giou_with_no_overlap(self):
+    expected_giou_tensor = [
+        0, -1/3, -3/4, 0, -98/100
+    ]
+
+    def graph_fn():
+      boxes1 = tf.constant([[3, 4, 5, 6], [3, 3, 5, 5],
+                            [0, 0, 0, 0], [3, 3, 5, 5],
+                            [9, 9, 10, 10]],
+                           dtype=tf.float32)
+      boxes2 = tf.constant([[3, 2, 5, 4], [3, 7, 5, 9],
+                            [5, 5, 10, 10], [3, 5, 5, 7],
+                            [0, 0, 1, 1]], dtype=tf.float32)
+
+      giou = ops.giou(boxes1, boxes2)
+      self.assertEqual(giou.dtype, tf.float32)
+
+      return giou
+
+    giou = self.execute(graph_fn, [])
+    self.assertAllClose(expected_giou_tensor, giou)
+
+  def test_giou_with_overlaps(self):
     expected_giou_tensor = [
-        1, 0, -1/3, 1/25, 1, -3/4
+        1/25, 1/4, 1/3, 1/7 - 2/9
     ]
 
     def graph_fn():
-      boxes1 = tf.constant([[3, 3, 5, 5], [3, 4, 5, 6],
-                            [3, 3, 5, 5], [2, 1, 7, 6],
-                            [1, 1, 1, 1], [0, 0, 0, 0]], dtype=tf.float32)
-      boxes2 = tf.constant([[3, 3, 5, 5], [3, 2, 5, 4],
-                                 [3, 7, 5, 9], [4, 3, 5, 4],
-                                 [1, 1, 1, 1], [5, 5, 10, 10]], dtype=tf.float32)
+      boxes1 = tf.constant([[2, 1, 7, 6], [2, 2, 4, 4],
+                            [2, 2, 4, 4], [2, 2, 4, 4]],
+                           dtype=tf.float32)
+      boxes2 = tf.constant([[4, 3, 5, 4], [3, 3, 4, 4],
+                            [2, 3, 4, 5], [3, 3, 5, 5]], dtype=tf.float32)
+
+      giou = ops.giou(boxes1, boxes2)
+      self.assertEqual(giou.dtype, tf.float32)
+
+      return giou
+
+    giou = self.execute(graph_fn, [])
+    self.assertAllClose(expected_giou_tensor, giou)
+
+  def test_giou_with_perfect_overlap(self):
+    expected_giou_tensor = [1]
+
+    def graph_fn():
+      boxes1 = tf.constant([[3, 3, 5, 5]], dtype=tf.float32)
+      boxes2 = tf.constant([[3, 3, 5, 5]], dtype=tf.float32)
+
+      giou = ops.giou(boxes1, boxes2)
+      self.assertEqual(giou.dtype, tf.float32)
+
+      return giou
+
+    giou = self.execute(graph_fn, [])
+    self.assertAllClose(expected_giou_tensor, giou)
+
+  def test_giou_with_zero_area_boxes(self):
+    expected_giou_tensor = [0]
+
+    def graph_fn():
+      boxes1 = tf.constant([[1, 1, 1, 1]], dtype=tf.float32)
+      boxes2 = tf.constant([[1, 1, 1, 1]], dtype=tf.float32)
+
+      giou = ops.giou(boxes1, boxes2)
+      self.assertEqual(giou.dtype, tf.float32)
+
+      return giou
+
+    giou = self.execute(graph_fn, [])
+    self.assertAllClose(expected_giou_tensor, giou)
+
+  def test_giou_different_with_l1_same(self):
+    expected_giou_tensor = [
+        2/3, 3/5
+    ]
+
+    def graph_fn():
+      boxes1 = tf.constant([[3, 3, 5, 5], [3, 3, 5, 5]], dtype=tf.float32)
+      boxes2 = tf.constant([[3, 2.5, 5, 5.5], [3, 2.5, 5, 4.5]],
+                           dtype=tf.float32)
 
       giou = ops.giou(boxes1, boxes2)
       self.assertEqual(giou.dtype, tf.float32)
@@ -1658,13 +1733,14 @@ class TestCoordinateConversion(test_case.TestCase):
 
   def test_coord_conv(self):
     expected_box_tensor = [
-        [0.5, 0.5, 5.5, 5.5], [2, 1, 4, 7]
+        [0.5, 0.5, 5.5, 5.5], [2, 1, 4, 7], [0, 0, 0, 0]
     ]
 
     def graph_fn():
-      boxes = tf.constant([[3, 3, 5, 5], [3, 4, 2, 6]], dtype=tf.float32)
+      boxes = tf.constant([[3, 3, 5, 5], [3, 4, 2, 6], [0, 0, 0, 0]],
+                          dtype=tf.float32)
 
-      converted = ops.cy_cx_h_w_to_ymin_xmin_ymax_xmax(boxes)
+      converted = ops.center_to_corner_coordinate(boxes)
       self.assertEqual(converted.dtype, tf.float32)
 
       return converted
@@ -1673,6 +1749,5 @@ class TestCoordinateConversion(test_case.TestCase):
     self.assertAllClose(expected_box_tensor, converted)
 
 
-
 if __name__ == '__main__':
   tf.test.main()