yolo11m-redbull-competitors

YOLO11m Red Bull competitors

1. monster
2. monster_can
3. celsius
4. celsius_can
5. sting
6. sting_can
7. alaninu
8. alaninu_can
9. white_monster_can

Dataset and training

Link to dataset: https://mediacatch.fra1.digitaloceanspaces.com/logo_recognition/data/redbull-competitors.zip.

Local copy: /datasets/logo/logos/redbull-competitors.

To train on a single RTX 3090:

yolo detect train data=data.yaml model=yolo11m.pt epochs=100 imgsz=1280 batch=8

and with four devices:

yolo detect train data=data.yaml model=yolo11m.pt epochs=100 imgsz=1280 batch=32 device=0,1,2,3

Usage

Using YOLO's own CLI:

yolo predict model=/path/to/best.pt source=/path/to/inputs conf=0.5

Using bifrost detector lib https://github.com/mediacatch/bifrost/tree/development/libs/detector:

uv run detect --conf-threshold 0.5 --imgsz 1280 /path/to/inputs redbull-competitors

Using HF:

from ultralytics import YOLO
from huggingface_hub import hf_hub_download

weights = hf_hub_download("MediaCatch/yolo11m-redbull-competitors", "best.pt")
model = YOLO(weights)
results = model("https://ultralytics.com/images/zidane.jpg", conf=0.5, imgsz=1280)

Labels

<img src="labels.jpg" alt="labels" width="75%"/>

Results

<img src="results.png" alt="results" width="75%"/> <img src="BoxPR_curve.png" alt="BoxPR_curve" width="50%"/> <img src="BoxP_curve.png" alt="BoxP_curve" width="50%"/> <img src="BoxR_curve.png" alt="BoxR_curve" width="50%"/> <img src="BoxF1_curve.png" alt="BoxF1_curve" width="50%"/> <img src="confusion_matrix_normalized.png" alt="confusion_matrix_normalized" width="50%"/>