How to dynamically adjust the maximum number of features extracted by the RTAB-Map front-end odometry?

[ScottishOwl]

I hope to dynamically adjust the maximum number of features extracted by the RTAB-Map front-end odometry (which I believe corresponds to the kVisMaxFeatures parameter) based on the uncertainty of the globally estimated pose (the trace of the covariance matrix) during the mapping process. Could you please advise which code file I should modify to achieve this?

Additionally, I attempted to modify this parameter directly in the mainloop() function of OdometryROS.cpp using the following code:
parameters_[Parameters::kVisMaxFeatures()] = "100";
this->updateParameters(parameters_);
However, this change did not take effect. During mapping, the output still shows similar logs, such as:
[rgbd_odometry-1] [INFO] [1772120446.888656658] [rtabmap.rgbd_odometry]: Odom: quality=389, std dev=0.068907m|0.021630rad, update time=0.025438s delay=0.330555s
The fact that the "quality" value is still bigger than 100 suggests that the parameter modification was not actually applied. How should I modify the parameters for the change to take effect correctly?

[matlabbe]

Unlike rtabmap node, we cannot change dynamically parameters of the odometry nodes, we can only set them on initialization. You would need to restart the node with different parameters.

ros2 run rtabmap_odom rgbd_odometry --ros-args -p Vis/MaxFeatures:='100' ...

[ScottishOwl]

Unlike rtabmap node, we cannot change dynamically parameters of the odometry nodes, we can only set them on initialization. You would need to restart the node with different parameters.

ros2 run rtabmap_odom rgbd_odometry --ros-args -p Vis/MaxFeatures:='100' ...

Very grateful for your reply. Thank you!
I've added a virtual void function in Odometry class and its child classes (i.e. OdometryF2M), in which I set Parameters[kVisMacFeatures] = "100" and then called the parser (i.e. regPipeline_->parseParameters(params) in OdometryF2M.cpp).
This seems to manage to implement the dynamic change of parameters.

[matlabbe]

Yes, that would be the way to do it. To implement that feature more broadly, we would need to limit to only a subset of parameters, but probably most Vis/ can work that way. I labelled that issue as "enhancement".