Flattened message structure and made use of CategoryDistribution

CMakeLists.txt

@@ -48,14 +48,14 @@ find_package(catkin REQUIRED COMPONENTS
 # Generate messages in the 'msg' folder
 add_message_files(
   FILES
+  CategoryDistribution.msg
+  ClassifierInfo.msg
   Classification2D.msg
   Classification3D.msg
-  ClassifierInfo.msg
   Detection2D.msg
   Detection2DArray.msg
   Detection3D.msg
   Detection3DArray.msg
-  BoundingBox3D.msg
 )
 
 ## Generate services in the 'srv' folder

README.md

@@ -18,6 +18,9 @@ primary types of classifiers:
   2. **Detectors**, which identify class probabilities as well as the poses of
   those classes given a sensor input
 
+The class probabilities are stored with a CategoryDistribution message, which
+is essentially a map from integer IDs to floats.
+
 Message types exist separately for 2D (using `sensor_msgs/Image`) and 3D (using
 `sensor_msgs\PointCloud2`). The metadata that is stored for each object is
 application-specific, and so this package places very few constraints on the

msg/Classification2D.msg

@@ -5,16 +5,9 @@
 
 Header header
 
-# IDs of elements detected with any nonzero probability.
-# Does not have to include all possible object
-# ids, the scores for any ids not listed are assumed to be 0.
-int64[] ids
-
-# The probabilities or confidence values of the detected elements. Must be
-# sorted to correspond with the ids field, above
-# 0-1, should sum to 1 or less, must be the same length as ids
-float64[] scores
+# Class probabilities
+CategoryDistribution results
 
 # The 2D data that generated these results (i.e. region proposal cropped out of
-# the image). Not required for all detectors, so it may be empty.
-sensor_msgs/Image source
+# the image). Not required for all use cases, so it may be empty.
+sensor_msgs/Image source_img

msg/Classification3D.msg

@@ -5,16 +5,9 @@
 
 Header header
 
-# IDs of elements detected with any nonzero probability.
-# Does not have to include all possible object
-# ids, the scores for any ids not listed are assumed to be 0.
-int64[] ids
+# Class probabilities
+CategoryDistribution results
 
-# The probabilities or confidence values of the detected elements. Must be
-# sorted to correspond with the ids field, above
-# 0-1, should sum to 1 or less, must be the same length as ids
-float64[] scores
-
-# The 2D data that generated these results (i.e. region proposal cropped out of
+# The 3D data that generated these results (i.e. region proposal cropped out of
 # the image). Not required for all detectors, so it may be empty.
-sensor_msgs/PointCloud2 source
+sensor_msgs/PointCloud2 source_cloud

msg/Detection2D.msg

@@ -1,16 +1,18 @@
-# Defines a 2D detection result.
+# Defines a 3D detection result.
 #
-# This extends a basic 2D classification by including position information,
+# This extends a basic 3D classification by including position information,
 # allowing a classification result for a specific position in an image to
 # to be located in the larger image.
 
-# non-geometric results of detection.
-vision_msgs/Classification2D classification
+Header header
+
+# Class probabilities
+CategoryDistribution results
 
 # The x/y position and (optional) rotation of the bounding box center.
 geometry_msgs/Pose2D pose
 
 # (Optional) The size of the bounding box surrounding the object. The center of
-# the bounding box is the position of the detection point.
+# the bounding box is the position of the pose.
 float32 bbox_size_x
 float32 bbox_size_y

msg/Detection3D.msg

@@ -4,8 +4,10 @@
 # allowing a classification result for a specific position in an image to
 # to be located in the larger image.
 
-# non-geometric class probability.
-vision_msgs/Classification3D classification
+Header header
+
+# Class probabilities
+CategoryDistribution results
 
 # The element's pose. This pose should be
 # defined as the pose of some fixed reference point on the object, such as the
@@ -14,5 +16,6 @@ vision_msgs/Classification3D classification
 # setting classification.header.frame_id.
 geometry_msgs/Pose pose
 
-# (Optional) The bounding box in 3D space that surrounds the object.
-vision_msgs/BoundingBox3D bbox
+# The 3D data that generated these results (i.e. region proposal cropped out of
+# the image). Not required for all detectors, so it may be empty.
+sensor_msgs/PointCloud2 source_cloud