How to install

conda create -y -n keras-jax python=3.10
conda activate keras-jax
pip install -r requirements-jax-cuda.txt

Change Backbone

Refer: https://keras.io/api/keras_cv/models/tasks/yolo_v8_detector/

# Create the model with mobilenet backbone
model = keras_cv.models.YOLOV8Detector.from_preset(
    num_classes=20,
    bounding_box_format="xywh",
    preset="mobilenet_v3_large",
)

# Can load pretrained weight if needed
model = keras_cv.models.YOLOV8Detector.from_preset(
    num_classes=20,
    bounding_box_format="xywh",
    preset="mobilenet_v3_large_imagenet",
    load_weights=True,
)

Run Example Train

python train.py

Full Tutorial with Default Yolov8 Backbone

Link: https://keras.io/examples/vision/yolov8/