Cleaned up the target_assigner and CenterNet meta arch such that the …

…depth

logic doesn't depend on the per_keypoint_offset value and use
get_batch_predictions_from_indices to be consistent with other prediction
heads.

PiperOrigin-RevId: 369925004

research/object_detection/core/target_assigner.py

@@ -1318,7 +1318,8 @@ def __init__(self,
                keypoint_std_dev=None,
                per_keypoint_offset=False,
                peak_radius=0,
-               compute_heatmap_sparse=False):
+               compute_heatmap_sparse=False,
+               per_keypoint_depth=False):
     """Initializes a CenterNet keypoints target assigner.
 
     Args:
@@ -1349,12 +1350,16 @@ def __init__(self,
         version of the Op that computes the heatmap. The sparse version scales
         better with number of keypoint types, but in some cases is known to
         cause an OOM error. See (b/170989061).
+      per_keypoint_depth: A bool indicates whether the model predicts the depth
+        of each keypoints in independent channels. Similar to
+        per_keypoint_offset but for the keypoint depth.
     """
 
     self._stride = stride
     self._class_id = class_id
     self._keypoint_indices = keypoint_indices
     self._per_keypoint_offset = per_keypoint_offset
+    self._per_keypoint_depth = per_keypoint_depth
     self._peak_radius = peak_radius
     self._compute_heatmap_sparse = compute_heatmap_sparse
     if keypoint_std_dev is None:
@@ -1686,14 +1691,15 @@ def assign_keypoints_depth_targets(self,
 
     Returns:
       batch_indices: an integer tensor of shape [num_total_instances, 3] (or
-        [num_total_instances, 4] if 'per_keypoint_offset' is set True) holding
+        [num_total_instances, 4] if 'per_keypoint_depth' is set True) holding
         the indices inside the predicted tensor which should be penalized. The
         first column indicates the index along the batch dimension and the
         second and third columns indicate the index along the y and x
         dimensions respectively. The fourth column corresponds to the channel
         dimension (if 'per_keypoint_offset' is set True).
-      batch_depths: a float tensor of shape [num_total_instances, 1] indicating
-        the target depth of each keypoint.
+      batch_depths: a float tensor of shape [num_total_instances, 1] (or
+        [num_total_instances, num_keypoints] if per_keypoint_depth is set True)
+        indicating the target depth of each keypoint.
       batch_weights: a float tensor of shape [num_total_instances] indicating
         the weight of each prediction.
       Note that num_total_instances = batch_size * num_instances *
@@ -1800,7 +1806,7 @@ def assign_keypoints_depth_targets(self,
       # Prepare the batch indices to be prepended.
       batch_index = tf.fill(
           [num_instances * num_keypoints * num_neighbors, 1], i)
-      if self._per_keypoint_offset:
+      if self._per_keypoint_depth:
         tiled_keypoint_types = self._get_keypoint_types(
             num_instances, num_keypoints, num_neighbors)
         batch_indices.append(

research/object_detection/core/target_assigner_test.py

@@ -1863,7 +1863,7 @@ def graph_fn():
           class_id=1,
           keypoint_indices=[0, 2],
           peak_radius=1,
-          per_keypoint_offset=True)
+          per_keypoint_depth=True)
       (indices, depths, weights) = cn_assigner.assign_keypoints_depth_targets(
           height=120,
           width=80,

research/object_detection/meta_architectures/center_net_meta_arch.py

@@ -2430,7 +2430,8 @@ def _initialize_target_assigners(self, stride, min_box_overlap_iou):
                 keypoint_std_dev=kp_params.keypoint_std_dev,
                 peak_radius=kp_params.offset_peak_radius,
                 per_keypoint_offset=kp_params.per_keypoint_offset,
-                compute_heatmap_sparse=self._compute_heatmap_sparse))
+                compute_heatmap_sparse=self._compute_heatmap_sparse,
+                per_keypoint_depth=kp_params.per_keypoint_depth))
     if self._mask_params is not None:
       target_assigners[SEGMENTATION_TASK] = (
           cn_assigner.CenterNetMaskTargetAssigner(stride))
@@ -2853,17 +2854,13 @@ def _compute_kp_depth_loss(self, input_height, input_width, task_name,
          gt_keypoint_depths_list=gt_keypoint_depths_list,
          gt_keypoint_depth_weights_list=gt_keypoint_depth_weights_list)
 
-    if kp_params.per_keypoint_offset and not kp_params.per_keypoint_depth:
-      batch_indices = batch_indices[:, 0:3]
-
     # Keypoint offset loss.
     loss = 0.0
     for prediction in depth_predictions:
-      # TODO(yuhuic): Update this function to use
-      # cn_assigner.get_batch_predictions_from_indices().
-      selected_depths = tf.gather_nd(prediction, batch_indices)
-      if kp_params.per_keypoint_offset and kp_params.per_keypoint_depth:
-        selected_depths = tf.expand_dims(selected_depths, axis=-1)
+      if kp_params.per_keypoint_depth:
+        prediction = tf.expand_dims(prediction, axis=-1)
+      selected_depths = cn_assigner.get_batch_predictions_from_indices(
+          prediction, batch_indices)
       # The dimensions passed are not as per the doc string but the loss
       # still computes the correct value.
       unweighted_loss = localization_loss_fn(