Skip to content

Quick_Start_Using_GUI

Jump to bottom
Viktor Seib edited this page Apr 30, 2021 · 8 revisions

Quick Start Using the GUI

This guide shows how to use the software to classify isolated point clouds.

  • Download example point cloud objects from the PCL repository
    (five files for training ism_train_*.pcd and five files for testing ism_test_*.pcd).
  • in the following steps I will assume that these files are downloaded and placed in the folder point-cloud-donkey/example_clouds.

For this example I will use the executable training_gui which is build from the code located in point-cloud-donkey/src/training_gui. Please note that the training_gui is not a full replacement for the command line interface of the eval_tool, but rather serves as means of visualization of various algorithm steps.

This is just a quick start example with the training_gui. For a detailed documentation please refer to Training GUI Tool Documentation.

Training

  • Run the following command from the point-cloud-donkey folder to start the application
./build/bin/training_gui
  • You will see an application window similar to this alt text
  • Click the button Load Train Config which is located on the right in the Training button group
  • This button corresponds to the -t argument of the command line interface
  • In the file dialog select the file point-cloud-donkey/config/qs_input_config.ism to load the config
  • In the next step we need to add point clouds and their labels for training
  • This corresponds to the -f argument of the command line interface
  • Click the button Add Cloud which is located on the right in the Training button group
  • In the opening dialog enter the class label ID in the text field
  • Then click Load Model and load the first training model from the point-cloud-donkey/example_clouds folder
  • Click OK to close the dialog to add a model
  • The selected point cloud will appear in the application window
  • Repeat loading the training objects for each object as outlined in the file data/qs_train_list.txt
  • These are the following point clouds with their respective class IDs:
example_clouds/ism_train_cat.pcd     0
example_clouds/ism_train_horse.pcd   1
example_clouds/ism_train_lioness.pcd 2
example_clouds/ism_train_michael.pcd 3
example_clouds/ism_train_wolf.pcd    4
  • After adding all objects, click Train ISM
  • When training is finished some bounding boxes will appear in the application window, ignore them for now ;)
  • Important: To proceed to using the training results you have to click Save ISM (this corresponds to the -o argument of the command line interface)
  • Let us assume the ISM is saved as quick_start_example.ism

Classification

  • If you have closed the application after training, run the following command from the point-cloud-donkey folder to start the application
./build/bin/training_gui
  • Click the button Load ISM which is located on the right in the Detection button group
  • Important: If you did not close the application after training, you still need to click Load ISM and select the trained model
  • Select the previously trained ISM model (e.g. quick_start_example.ism)
  • This button corresponds to the -d argument of the command line interface
  • Each file for classification has to be loaded manually using the button Load Scene
  • You can try each of the files listed in the file data/qs_test_list.txt
  • These are the following point clouds with their respective ground truth class IDs:
example_clouds/ism_test_cat.pcd     0
example_clouds/ism_test_horse.pcd   1
example_clouds/ism_test_lioness.pcd 2
example_clouds/ism_test_michael.pcd 3
example_clouds/ism_test_wolf.pcd    4
  • Selecting any of these files and clicking Detect should yield the given ground truth class ID as result
  • Important: the class ID is only displayed on the command line, but not in the application window

Classification Example

  • Let us assume we have loaded the previously trained ISM model
  • For the sake of this example, make sure that all checkboxes on the lower right are cleared except for Show Results
  • We proceed by clicking Load Scene and selecting the file example_clouds/ism_test_cat.pcd
  • The point cloud of the cat will appear in the application window (you might need to zoom out using the mouse wheel) alt text
  • Now click Detect and many different bounding boxes will appear in the window
  • Meanwhile you should see an output similar to this in the console:
[13:03:13] INFO: finding maxima
[13:03:13] INFO: maximum 0, class: 0, weight: 0.705494, instance: 0 (0.705494), num votes: 13
[13:03:13] INFO: maximum 1, class: 1, weight: 0.116802, instance: 1 (0.116802), num votes: 2
[13:03:13] INFO: maximum 2, class: 1, weight: 0.0592346, instance: 1 (0.0592346), num votes: 1
[13:03:13] INFO: maximum 3, class: 3, weight: 0.0592346, instance: 3 (0.0592346), num votes: 1
[13:03:13] INFO: maximum 4, class: 4, weight: 0.0592346, instance: 4 (0.0592346), num votes: 1
[13:03:13] INFO: detected 5 maxima
  • These are all detection hypotheses in descending order of confidence
  • You will note that class ID 0 is the top result which shows that the cat was classified correctly
  • Also note that the top result (cat) has 13 votes, while the second result has only 2
  • Next enter any number between 3 and 13 in the text field labeled Min Votes: in the lower right corner
  • After clicking the Detect button again the visualization is much nicer alt text
  • You can proceed by using other point clouds for classification
Clone this wiki locally