Skip to content

Latest commit

 

History

History
83 lines (66 loc) · 2.38 KB

README.md

File metadata and controls

83 lines (66 loc) · 2.38 KB
Raw

Clone this Repo

git clone {https://github.com/tannd-ds/this_repo.git} {REPO_HOME}

YOLOv8

Setup Environment, please replace {ENV_NAME} with name of your choice.

conda create -n {ENV_NAME} python=3.10
conda activate {ENV_NAME}
pip install -r requirements.txt

To run the demo of YOLO on VisDrone dataset (test set)

python {REPO_HOME}/run.py \
    --MODEL yolo \
    --WEIGHTS_PATH {path_to_yolo_weights} \
    --SEQUENCES_DIR {path_to_visdrone_dataset_test/test} \
    --TRACKER botsort \
    --SAVE_RESULTS \
    --SHOW

YOLOv10

Note: YOLOv10 is not officially part of Ultralytics, we need to set up it differently.

Clone YOLOv10 Repo

git clone https://github.com/NhiNguyen34/yolov10.git {YOLOv10_HOME}
cp {REPO_HOME}/my_utils/run_yolov10_on_visdrone.py {YOLOv10_HOME}

Note: You need to copy the {REPO_HOME}/my_utils/run_yolov10_on_visdrone.py to {YOLOv10_HOME} since it need to use the ultralytics of YOLOv10 instead of the official one.

Setup Environment, please replace {ENV_NAME} with name of your choice.

cd {YOLOv10_HOME}
conda create -n {ENV_NAME} python=3.10
conda activate {ENV_NAME}
pip install -r requirements.txt

To run the demo of YOLO on VisDrone dataset (test set)

cd {YOLOv10_HOME}
python run_yolov10_on_visdrone.py \
    --WEIGHTS_PATH {path_to_yolov10_weights} \
    --SEQUENCES_DIR {path_to_visdrone_dataset_test/test} \
    --TRACKER botsort \
    --SAVE_RESULTS \
    --SHOW

Evaluation

We use TrackEval for evaluation. To run the evaluation:

cd {REPO_HOME}

python TrackEval/scripts/run_visdrone.py \
    --BENCHMARK VisDrone2019-MOT_coco \
    --DO_PREPROC False \
    --SPLIT_TO_EVAL [train/test] \
    --TRACKERS_TO_EVAL {TRACKERS_NAME} \
    --USE_PARALLEL True

Where to find your results?

  • After you run above code (the run.py, in previous section), by default, your results will be save to TrackEval/data/trackers/mot-challenge/VisDrone2019-MOT_coco/{SPLIT}/{TRACKERS_NAME}. SPLIT will be train or test based on your previous run on VisDrone train or test set respectively.

Acknowledgement

Some parts of our code are borrowed from the following works: