├── demo_pic_and_video
├── ROAD_Waymo_Baseline
├── Road-waymo-dataset
├── runs (Save model's weight)
│ ├── action
│ ├── detect
│ └── location
├── Track1
├── Track2
├── utils
conda create --name ROADpp python>=3.10
conda activate ROADpp
pip install -r requirement.txt
├── runs (Save model's weight)
│ ├── action
│ ├── detect
│ └── location
You need to first understand how to configure the YAML file for YOLOv8, as well as the dataset format.
cd ROADpp_challenge_ICCV2023
python detect.py --video_path 'xxx' --yolo_path 'xxx' --two_branch True --major_path 'xxx' rare_path 'xxx' --pkl_name 'xxx' --save_res 'xxx'
mode
: Detect mode, only accepts Track1 or Track2.
video_path
: Path to the video.
yolo_path
: Path to the YOLO model.
two_branch
: Indicates whether to use two-branch YOLO.
major_path
: Path to the major YOLO model.
rare_path
: Path to the rare YOLO model.
devices
: GPU number.
imgsz
: YOLO input size.
video_shape
: Original video resolution.
submit_shape
: Final submit shape.
pkl_name
: Submit file name (*.pkl).
save_res
: Save submit file.
action_detector_path
: Path to the action detector model (Track2 only).
loc_detector_path
: Path to the location detector model (Track2 only).
t2_input_shape
: Track 2 input shape.
windows_size
: Sliding windows shape.