how can I train a classifier with my data.

[szm88]

@rdube Hi,rdube. now I have laser data in real environment,and i want to train the model.how can i train a classifier?

[Durant35]

• It seems that there are some rf_xxx parameters needed in training step
• In my opinion, you can do as following
  • [option] Segment to get cloud segments of your desired objects
  • calculate those segments' features using Eigenvalue-based Features EigenvalueBasedDescriptor or Ensemble of Shapes Functions EnsembleShapeFunctions
  • Then you can train the OpenCvRandomForest, parameters are needed???

    void OpenCvRandomForest::train(const Eigen::MatrixXd& features,
                           const Eigen::MatrixXd& labels)
    {
    ...

[rdube]

Hi guys, Unfortunately I do not have the time to support training a RF for segmatch. I can try to find the parameters that we used for training if you really do need them.

For your information, we will update segmatch soon and will not be supporting random forests anymore. The issue is that they can be too expensive to use when doing localization in large maps, in comparison to doing k-NN retrieval in feature space. With the upcoming version we can do localization at 10Hz on velodyne data. See our RA-L paper on this strategy here: incremental localization

We will soon include a new descriptor which might be of interest to you. In the mean-time I would encourage you to use the eigen-value based descriptor with knn retrieval. We will update the code and demonstrations soon and will be happy to help you using this descriptor if you have any issue. Hope that helps!

[Durant35]

Great！Thanks a lot for your kindly replay. I will try your suggestions soon. Hope to feedback.

[rdube]

@Durant35 Here is a video demonstration of the data-driven descriptor I was mentioning. We are now updating the approach and won't be using RF anymore (see #79). Please feel free to reopen an issue if you have further questions!

[Durant35]

Awesome！Thanks for your kindly reply.