can't load the saved model YOLOV8Backbone from keras_cv after thefine-tuning

[DavidMacha]

I’m trying to fine-tune and save a YOLOv8 model using keras_cv.models.YOLOV8Detector, but I keep running into checkpoint and model loading issues. Below is the setup and errors I’ve encountered — I’d really appreciate any insights or fixes!

checkpoint_filepath='E:\model.yolo_v8_s_ft.h5'
# backbone = keras_cv.models.YOLOV8Backbone.from_preset("yolo_v8_m_backbone_coco") 
backbone = keras.models.load_model(checkpoint_filepath) 

yolo = keras_cv.models.YOLOV8Detector(
    num_classes=len(class_mapping),
    bounding_box_format="xyxy",
    backbone=backbone,
    fpn_depth=2,)
optimizer = tf.keras.optimizers.Adam(learning_rate=LEARNING_RATE,global_clipnorm=GLOBAL_CLIPNORM,)
yolo.compile(optimizer=optimizer, classification_loss="binary_crossentropy", box_loss="ciou")


Save_mode= keras.callbacks.ModelCheckpoint(filepath=checkpoint_filepath,monitor='val_loss',mode='auto',save_best_only=True,save_freq="epoch")
reduce_lr = keras.callbacks.ReduceLROnPlateau(monitor='val_loss', factor=0.5,patience=3, min_lr=0.0001)
Stop=keras.callbacks.EarlyStopping(monitor="val_loss",patience=7,mode="auto")

yolo.fit(
    train_ds,
    validation_data=val_ds,
    epochs=EPOCH,
    callbacks=[Save_mode, reduce_lr, Stop],
)

---

❌ Error 1: Unknown Layer

ValueError: Unknown layer: 'YOLOV8Detector'

---

🧩 Attempted Fix #1 — Using TensorFlow Hub

I tried specifying custom objects when loading:

import tensorflow_hub as hub

backbone = keras.models.load_model(
    checkpoint_filepath,
    custom_objects={'YOLOV8Detector': hub.YOLOV8Detector}
)

But this gives another error:

AttributeError: module 'tensorflow_hub' has no attribute 'YOLOV8Detector'

---

🧩 Attempted Fix #2 — Saving as .tf

I also tried changing the model checkpoint extension:

checkpoint_filepath = r"E:\model.yolo_v8_s_ft.tf"

However, this raised:

ValueError: The filepath provided must end in .keras (Keras model format). 
Received: filepath=E:\model.yolo_v8_s_ft.tf

---

[Mattral]

Possible Solution (Saw This on Stack Overflow)

I came across a similar issue before, this error usually happens because Keras doesn’t recognize the YOLOV8Detector class when loading the model.

The main problem is that the entire detector model was saved, but it’s being loaded as a backbone instead. To fix it, you can explicitly define the custom object during loading.

Here’s how you can do it properly:

#  Save the entire detector model
checkpoint_filepath = 'E:/model.yolo_v8_s_ft.keras'
yolo.save(checkpoint_filepath)

#  Later, load with custom objects
from keras_cv.models.object_detection.yolo_v8 import YOLOV8Detector

loaded_model = keras.models.load_model(
    checkpoint_filepath,
    custom_objects={'YOLOV8Detector': YOLOV8Detector}
)

This ensures that Keras knows how to interpret the YOLOV8Detector class when restoring the model from disk.