Noetic support about dnn_detect

[davecrawley]

Thank you for your PR - that's the perfect thing to do. We are expecting to do a noetic release shortly, probably in the next few months.

[tim-fan]

Ok great, thanks for the info. I look forward to the release!

[tim-fan]

Re-opening to make some more comments;

Thanks for merging #7 . While I was able to build and run on noetic with those changes, I've taken a closer look at my dockerfile, and noticed I was using the -r flag for rosdep install, which ignores errors when resolving dependencies. This was masking issues with two dependencies listed in package.xml:

1. fiducial_msgs this has not been released for Noetic, but on the other hand I don't see any use of this package in dnn_detect, so I recommend just deleting that line. I've opened #9 for this change.
2. opencv3 - this is not available in noetic. Reading up on how opencv versions have been supported from kinetic to noetic, my understanding is that the opencv3 package was created to allow kinetic packages to use opencv3.3.1. This was required for dnn_detect in kinetic in order to use the opencv dnn module. In noetic the system installed opencv version 4.2 which includes the dnn module, so it looks like opencv3 is no longer required, and can be switched to libopencv-dev.

Note change (2) will break kinetic, so I assume the change should be made on a noetic-devel branch. I've made the change in https://github.com/tim-fan/dnn_detect/tree/noetic-devel, but I believe you need to create a noetic-devel branch before I can pull request the change to this repo. So if you'd like to proceed with the change, let me know when the branch is created and I'll open the pull request.

References:
Discussion on opencv version support for melodic: ros-infrastructure/rep#167
Supported opencv version for noetic: https://www.ros.org/reps/rep-0003.html#noetic-ninjemys-may-2020-may-2025

[rohbotics]

Thanks for looking into this, I merged #9. As for a noetic branch, I am not too excited about branching development for such a small package, is it possible that we don't have to directly depend on opencv and have it bring in the correct version automatically by depending on cv_bridge?

If that doesn't seem to work, I am ok with creating a noetic branch.

[tim-fan]

Hey, nice idea, I've given that a test (just remove the opencv3 dep), and it works on my docker builds for both kinetic and noetic - the dependency on cv_bridge pulls in a working version of opencv in both cases. So it looks like a new noetic-devel branch wont be needed.

On the one hand I imagine it would preferable to be explicit about such dependencies, rather than relying on other dependencies to pull in the correct thing. However in this case I guess that is outweighed by the desire to keep things simple (single devel branch). I'll create the pull request for this change now.

[rohbotics]

@tim-fan I merged your PR. Let me know if everything is working on Noetic now, I will do a release of the package.

Thanks for all the help.

[tim-fan]

Last change (hopefully!) : #11
That came up when I was trying to build/run the test for noetic.
Once that's resolved I'll do a final test for the release.

[tim-fan]

Ok, with that change merged, I used the dockerfile from here to confirm that the kinetic-devel branch builds with noetic, tests pass, and that I can run dnn_detect.launch and get detected objects (confirmed I am a person sitting on a sofa 😄 )

So looks good to me for a noetic release 👍

[Alex-Beh]

Hello, can I use this dnn_detect with opencv_3.4? I try use the package with the following two command:
roslaunch usb_cam usb_cam-test.launch
roslaunch dnn_detect dnn_detect.launch camera:=/usb_cam image:=image_raw

And it raised the following error:

cv exception: /build/opencv-L2vuMj/opencv-3.2.0+dfsg/modules/core/src/matrix.cpp:1032: error: (-215) dims <= 2 in function reshape

I try to print out the shape of the cv_ptr->image that pass into resize function and it is [640,480]. The package can only be used on compressed image as input?

---

Update: Solved by git clone the cv_bridge and checkout the melodic branch.