Skip to content

Latest commit

 

History

History
111 lines (82 loc) · 4.32 KB

README.md

File metadata and controls

111 lines (82 loc) · 4.32 KB
Raw

Data Preparation(KITTI dataset)

Weak Supervision

For a given KITTI RAW sequence, use your preferred 2D segmentation method(we made use of inplace-abn) to generate the results for the corresponding sequence and store it following the same file hierarchy. Pre-generated segmentation using inplace_abn is provided here. Preprocessed weak static layouts for KITTI RAW dataset can be downloaded from here.

.data/raw/
└── 2011_09_26
    ├── 2011_09_26_drive_0001_sync
    │   ├── oxts
    │   │   ├── data
    │   │   │   ├── 0000000000.txt
    │   │   │   ├── 0000000001.txt
    │   │   │   ├── 0000000002.txt
    │   │   │   ├── ........
    │   │   ├── dataformat.txt
    │   │   ├── oxts
    │   │   └── timestamps.txt
    │   ├── segmentation
    │   │   ├── 0000000000.png
    │   │   ├── 0000000001.png
    │   │   ├── 0000000002.png
    │   │   ├── ..........
    │   └── velodyne_points
    │       ├── data
    │       │   ├── 0000000000.bin
    │       │   ├── 0000000001.bin
    │       │   ├── ........
    │       ├── timestamps_end.txt
    │       ├── timestamps_start.txt
    │       └── timestamps.txt
    ├── calib_cam_to_cam.txt
    ├── calib_imu_to_velo.txt
    └── calib_velo_to_cam.txt

python3 generate_supervision.py --base_path <path to kitti raw root directory> --date <date of recorded KITTI sequence> --sequence <sequence number> --range <range of the rectangular layout (in m)> --occ_map_size <size of the rectangular occupancy map (in px)> --seg_class <road/sidewalk>

For eg.

python3 generate_supervision.py --base_path ../../data/raw/ --seg_class road --date 2011_09_26 --sequence 0001 --range 40 --occ_map_size 256

For obtaininig static layouts for entire KITTI RAW dataset run:

python3 generate_suprevision.py --base_path <path to kitti raw root directory> --seg_class road --process all

GroundTruth

Groundtruth static layouts for KITTI RAW and KITTI Odometry datasets are manually annotated in bird's eye view and can be downloaded from here. For generating groundtruth as well as training data for KITTI 3Dobject one should organize the data in the following hierarchy.

data/object/
├── training
│   ├── calib
│   ├── image_2
│   ├── label_2
├── testing
    └── calib




python3 generate_supervision.py --base_path ./data/object/training/label_2 --seg_class vehicle --range 40 --occ_map_size 256

Usage

usage: generate_supervision.py [-h] [--base_path BASE_PATH] [--date DATE]
                               [--sequence SEQUENCE] [--out_dir OUT_DIR]
                               [--range RANGE] [--occ_map_size OCC_MAP_SIZE]
                               [--seg_class {road,sidewalk,vehicle}]
                               [--process {all,one}]

MonoLayout DataPreparation options

optional arguments:
  -h, --help            show this help message and exit
  --base_path BASE_PATH
                        Path to the root data directory
  --date DATE           Corresponding date from the KITTI RAW dataset
  --sequence SEQUENCE   Sequence number corresponding to a particular date
  --out_dir OUT_DIR     Output directory to save layouts
  --range RANGE         Size of the rectangular grid in metric space (in m)
  --occ_map_size OCC_MAP_SIZE
                        Occupancy map size (in pixels)
  --seg_class {road,sidewalk,vehicle}
                        Data Preparation for Road/Sidewalk/Vehicle
  --process {all,one}   Process entire KITTI RAW dataset or one sequence at a
                        time