Depth image encoding is hard-coded

[gavanderhoorn]

The current implementation appears to hard-code the expected image encoding of the depth image:

industrial_reconstruction/industrial_reconstruction/industrial_reconstruction/industrial_reconstruction.py

Line 288 in 2cf137d

cv2_depth_img = self.bridge.imgmsg_to_cv2(depth_image_msg, "16UC1")

it doesn't do this for the regular RGB images:

industrial_reconstruction/industrial_reconstruction/industrial_reconstruction/industrial_reconstruction.py

Line 289 in 2cf137d

cv2_rgb_img = self.bridge.imgmsg_to_cv2(rgb_image_msg, rgb_image_msg.encoding)

Looking at various drivers, there are two main 'types' of depth images:

• 16-bit unsigned with depth in mm (16UC1)
• 32-bit float with depth in m (32FC1)

At least in ROS 1, the depth_image_proc/convert_metric nodelet can be used to convert between the two types.

The hard-coded 16UC1 (and the existence of the depth_scale parameter) implies the first is expected and has been used when developing industrial_reconstruction. The code as-is is incompatible with the second type.

Would it be desirable to remove the hard-coding and rely on what is set in sensor_msgs::msgs::Image::encoding field instead?

[marrts]

Would it be desirable to remove the hard-coding and rely on what is set in sensor_msgs::msgs::Image::encoding field instead?

Yes, this would definitely be preferable. Upon initial creation of this node I had some issues trying to do that, so I just ran with what would work for the initial application. It sounds like you and @aashish-tud have an application using different camera encodings than what I've used in the past, if you guys have a working change for that I would definitely want to merge that in.