search

ros-perception / vision_msgs

Algorithm-agnostic computer vision message types for ROS.
Apache License 2.0
149 stars 72 forks source link

Improper use of Pose2D datatype in BoundingBox2D #65

Closed tfoote closed 2 years ago

tfoote commented 2 years ago

The semantic definition of Pose2D datatype represents metric position in meters (As per the default units REP 103) while the BoundingBox2D says that it's in pixels.

https://github.com/ros-perception/vision_msgs/blob/ef04cafde553780569bb53f0d0bd3d0edc04bfc4/msg/BoundingBox2D.msg#L7-L8

This should not be redefining the units in an established submessage. And furthermore the Pose2D message is deprecated thus should not be being used.

A pixel space representation of this should be created. It could have the same payload. And even the same message name in vision_msgs namespace. Ala vision_msgs/Pose2D.msg but with the units appropriately defined as pixels instead of meters.

Also the axes and direction of the angle should be defined. There are a couple common but different conventions in image space.

SteveMacenski commented 2 years ago

Seems reasonable to me. @ijnek can you update your PR in guidance with this instead? https://github.com/ros-perception/vision_msgs/pull/64