Skip to content

Latest commit

 

History

History
215 lines (157 loc) · 4.75 KB

3_eval.md

File metadata and controls

215 lines (157 loc) · 4.75 KB

Evaluation

Configuring rosbag data for data conversion

prepare the data and dependency for code

# download the rosbag and put them under rosbag
cd path/to/crowdbot_tools/data
mkdir rosbag && cd rosbag
# (recommended) create symbolic links of rosbag folder instead of copying data
# ref: https://stackoverflow.com/questions/9587445/how-to-create-a-link-to-a-directory
ln -s /hdd/data_qolo/lausanne_2021/nocam_rosbags/
ln -s  /hdd/data_qolo/lausanne_2021/24_04_2021/shared_control 0424_shared_control
ln -s  /hdd/data_qolo/lausanne_2021/24_04_2021/RDS/detector 0424_rds_detector

# install tf2-sensor-msgs to get transform functions
sudo apt-get install ros-$ROS_DISTRO-tf2-sensor-msgs

Extracting source data from rosbag

Pointcloud & LiDAR Timestamp

python3 qolo/gen_lidar_from_rosbags.py -f 0424_shared_control
  • input:
  • output:
  • detailed description:

tf_qolo information

python3 qolo/tfqolo2npy.py -f 0424_shared_control
  • input:
  • output:
  • detailed description: TODO: include the method to filter the signal

twist information

python3 qolo/twist2npy.py -f 0424_shared_control
  • input:
  • output:
  • detailed description:

pose information

python3 qolo/pose2d2npy.py -f 0424_shared_control
  • input:
  • output:
  • detailed description:

command information

python3 qolo/commands2npy.py -f 0424_shared_control
  • input:
  • output:
  • detailed description:

Running detector with Person-MinkUNet

  1. prepare checkpoints: downloads checkpoints into qolo/Person_MinkUNet/models for erson_MinkUNet

  • Person_MinkUNet/models structure

    ├── models
│   ├── ckpt_e40_train.pth
│   ├── ckpt_e40_train_val.pth
│   ├── unet_bl_voxel_jrdb_0.05_0.1_train_vel.yaml
│   └── unet_bl_voxel_jrdb_0.05_0.1_train.yaml
  1. run the detection code
python3 qolo/gen_detection_res.py -f 0424_shared_control
  • input:
  • output:
  • detailed description:

Running tracker with AB3DMOT

  1. add PYTHONPATH variable (optional)

    As indicated here, to load the library appropriately, please remember to append PYTHONPATH variable in each terminal or add the following to your ~/.profile

    # conda activate torch38_cu110
export PYTHONPATH="${PYTHONPATH}:${PWD}/qolo/AB3DMOT"
echo $PYTHONPATH

    Inspired by these two stackoverflow posts, [In Python script, how do I set PYTHONPATH?] and [How do I get the full path of the current file's directory?], we can append the AB3DMOT library in python scripts as follows:

    import os
import sys
# export PYTHONPATH="${PYTHONPATH}:${PWD}/qolo/AB3DMOT"
sys.path.append(os.path.join(os.path.dirname(os.path.abspath(__file__)), "AB3DMOT"))

  2. run the tracking code

python3 qolo/gen_tracking_res.py -f 0424_shared_control
  • input:
  • output:
  • detailed description:

Visualizing scans, detections, and tracking results

viz with Open3D

  • running scripts
python3 qolo/gen_viz_img.py -f 0424_shared_control

video & gifs generated from image sequence

python3 qolo/gen_animation.py -f 0424_shared_control

Metrics & crowd characteristics

Path efficiency-related metrics

python3 qolo/eval_qolo_path.py -f 0424_shared_control
  • input:
  • output:
  • parameters: (in table way)
  • detailed description:

Qolo control performance-related metrics

python3 qolo/eval_qolo_ctrl.py -f 0424_shared_control
  • input:
  • output:
  • parameters: (in table way)
  • detailed description:

Crowd-related metrics

python3 qolo/eval_crowd.py -f 0424_shared_control
  • input:
  • output:
  • parameters: (in table way)
  • detailed description:

Other utility functions

Dataset structure

taking 0424_mds_processed/ as an example

$ tree -L 2
.
├── alg_res
│   ├── detections
│   └── tracks
├── lidars
│   ├── 2021-04-24-12-04-04
│   ├── 2021-04-24-12-07-57
│   ├── 2021-04-24-12-10-45
│   ├── 2021-04-24-12-54-04
│   ├── 2021-04-24-12-56-59
│   └── 2021-04-24-13-03-39
├── metrics
│   ├── 2021-04-24-12-04-04
│   ├── 2021-04-24-12-07-57
│   ├── 2021-04-24-12-10-45
│   ├── 2021-04-24-12-54-04
│   ├── 2021-04-24-12-56-59
│   └── 2021-04-24-13-03-39
└── source_data
    ├── commands
    ├── pose2d
    ├── tf_qolo
    ├── timestamp
    └── twist