Python KittiViewer+

This is an upgrade of Second.KittiViewer

• Works on raw KITTI data (without second preprocessing)
• Allows to compare results of different methods (results must be saved in KITTI format)
• Allows to display segmentation of points

Install Dependencies

pip install numba dill matplotlib fire shapely scikit-image PyOpenGL protobuf

It's easier to use conda to get QT:

conda install qt pyqt pyqtgraph

QT errors

• If you have “Failed to load platform plugin ”xcb“ ” while launching qt5 app on linux without qt installed error, it is probably because libxcb-xinerama0 file is not found. You can try run sudo apt install --reinstall libxcb-xinerama0 to fix it.

Run

python viewer.py

Parameters

All parameters are stored in params.json.

KITTI Detection

Data setup

• kitti_detection_root parameter contains a path for detection KITTI dataset. You must have image_2, label_2, velodyne, calib folders inside "selected_split": "training" folder.

  "kitti_detection_root": "/data/sets/kitti_second"

• folders parameter contains names for each modality (if you named them differently from KITTI).

  "folders": {
      "detection": {
          "calib": "calib",
          "image": "image_2",
          "label": "label_2",
          "velodyne": "velodyne"
      },
  }

• compare_results_detection object describes how to display results from different methods. All results must be stored inside the root folder and methods[i]["name"] subfolder in KITTI annotations format (.txt files as in label_2). color field describes RGB color of 3d and 2d boxes.

      "compare_results_detection": {
  
          "root": "/data/sets/kitti_results_detection",
          "methods": [
  
              {
                  "name": "method1",
                  "color": [1.0, 0.0, 1.0]
              },
              {
                  "name": "method2",
                  "color": [1.0, 1.0, 0.0]
              },
              {
                  "name": "method3",
                  "color": [0.0, 0.0, 1.0]
              }
          ]
      }

Example

KITTI Tracking

Data setup

• kitti_tracking_root parameter contains a path for detection KITTI dataset. You must have image_2, label_2, velodyne, calib folders inside "selected_split": "training" folder. image_2 and velodyne must have folders for each scene.

  "kitti_tracking_root": "/data/sets/kitti_tracking"

• folders parameter contains names for each modality (if you named them differently from KITTI).

  "folders": {
      "tracking": {
          "calib": "calib",
          "image": "image_2",
          "label": "label_2",
          "velodyne": "velodyne"
      },
  }

• compare_results_tracking object describes how to display tracking results from different methods. All results must be stored inside the root folder and methods[i]["name"] subfolder in selected format. color field describes RGB color of 3d and 2d boxes.

      "compare_results_detection": {
  
          "root": "/data/sets/kitti_results_detection",
          "format": "ab3dmot",
          "methods": [
  
              {
                  "name": "method1",
                  "color": [1.0, 0.0, 1.0]
              },
              {
                  "name": "method2",
                  "color": [1.0, 1.0, 0.0]
              },
              {
                  "name": "method3",
                  "color": [0.0, 0.0, 1.0]
              }
          ]
      }

• Supported tracking annotations formats:

  1. ab3dmot: comma separated values in order:

Frame	Type	2D BBOX (x1, y1, x2, y2)	Score	3D BBOX (h, w, l, x, y, z, rot_y)	Alpha
0	2 (car)	726.4, 173.69, 917.5, 315.1	13.85	1.56, 1.58, 3.48, 2.57, 1.57, 9.72, -1.56	-1.82

Types: {1:'Pedestrian', 2:'Car', 3:'Cyclist'}

Example

KITTI Segmentation (In development)

Data setup

• kitti_segmentation_root parameter contains a path for detection KITTI dataset. You must have .bin files "selected_split": "training" folder. Each file is created using dill library and must contain segmented_points (NxC, C - amount of classes (currently C=1 only),original_points (Nx3 - with points coordinates), ground_trurh_points (NxC - with ground truth segmentation)

  "kitti_segmentation_root": "/data/sets/kitti_segmentation"

Example