KerasCV YOLO模型保存加载后报错：'_DictWrapper'对象不可调用

参考Keras官方YOLOv8教程构建目标检测模型，训练完成后保存模型，加载后调用visualize_detections函数时触发TypeError，提示'_DictWrapper' object is not callable，未执行保存加载操作时该函数可正常运行。

复现代码 #

backbone = keras_cv.models.YOLOV8Backbone.from_preset("yolo_v8_xs_backbone_coco")
yolo = keras_cv.models.YOLOV8Detector(
    num_classes=len(class_mapping),
    bounding_box_format="xyxy",
    backbone=backbone,
    fpn_depth=1,
)
optimizer = tf.keras.optimizers.Adam(
    learning_rate=LEARNING_RATE,
    global_clipnorm=GLOBAL_CLIPNORM,
)

yolo.compile(
    optimizer=optimizer, classification_loss="binary_crossentropy", box_loss="ciou"
)


stopping_patience = 15
stopping_delta = 0.01
early_stopping = tf.keras.callbacks.EarlyStopping(monitor='val_loss',
                                                  patience=stopping_patience,
                                                  min_delta=stopping_delta,
                                                  verbose=1,
                                                  restore_best_weights=True)

history = yolo.fit(
    train_ds,
    validation_data=val_ds,
    epochs=EPOCH,
    callbacks=[EvaluateCOCOMetricsCallback(val_ds, "model_v2"), early_stopping],
)
# saving
yolo.save("myyolov8model.keras")
# loading
model = load_model("myyolov8model_withtraindata_v3.keras",
            custom_objects={'YOLOV8Detector': keras_cv.models.YOLOV8Detector}, compile=False)
model.compile(
        optimizer=optimizer, classification_loss="binary_crossentropy", box_loss="ciou")
visualize_detections(model, dataset=val_ds, bounding_box_format="xyxy", class_mapping=class_mapping)

报错详情 #

TypeError:   File "C:\Users\marlauwe\AppData\Local\anaconda3\Lib\site-packages\keras_cv\models\object_detection\yolo_v8\yolo_v8_detector.py", line 609, in decode_predictions
    return self.prediction_decoder(box_preds, scores)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
TypeError: in user code:

    File "C:\Users\marlauwe\AppData\Local\anaconda3\Lib\site-packages\keras\src\engine\training.py", line 2416, in predict_function  *
        return step_function(self, iterator)
    File "C:\Users\marlauwe\AppData\Local\anaconda3\Lib\site-packages\keras\src\engine\training.py", line 2401, in step_function  **
        outputs = model.distribute_strategy.run(run_step, args=(data,))
    File "C:\Users\marlauwe\AppData\Local\anaconda3\Lib\site-packages\keras\src\engine\training.py", line 2389, in run_step  **
        outputs = model.predict_step(data)
    File "C:\Users\marlauwe\AppData\Local\anaconda3\Lib\site-packages\keras_cv\models\object_detection\yolo_v8\yolo_v8_detector.py", line 616, in predict_step
        return self.decode_predictions(outputs, args[-1])
    File "C:\Users\marlauwe\AppData\Local\anaconda3\Lib\site-packages\keras_cv\models\object_detection\yolo_v8\yolo_v8_detector.py", line 609, in decode_predictions
        return self.prediction_decoder(box_preds, scores)

    TypeError: '_DictWrapper' object is not callable

解决方案 #

问题根源 #

模型保存与加载过程中，YOLOV8Detector的prediction_decoder组件未被正确序列化，加载后变成了_DictWrapper配置包装对象，而非可调用的解码器实例，导致调用时触发类型错误。

修复步骤 #

加载模型后，手动重新实例化YOLOV8PredictionDecoder并赋值给模型的prediction_decoder属性：

# 加载模型后添加以下代码
from keras_cv.models.object_detection.yolo_v8 import YOLOV8PredictionDecoder

model.prediction_decoder = YOLOV8PredictionDecoder(
    bounding_box_format="xyxy",
    num_classes=len(class_mapping),
    score_threshold=0.25,  # 可根据需求调整，默认0.25
    iou_threshold=0.7       # 可根据需求调整，默认0.7
)

修改后完整加载代码片段 #

# loading
model = load_model("myyolov8model_withtraindata_v3.keras",
            custom_objects={'YOLOV8Detector': keras_cv.models.YOLOV8Detector}, compile=False)
# 重新设置prediction_decoder
from keras_cv.models.object_detection.yolo_v8 import YOLOV8PredictionDecoder
model.prediction_decoder = YOLOV8PredictionDecoder(
    bounding_box_format="xyxy",
    num_classes=len(class_mapping),
    score_threshold=0.25,
    iou_threshold=0.7
)
model.compile(
        optimizer=optimizer, classification_loss="binary_crossentropy", box_loss="ciou")
visualize_detections(model, dataset=val_ds, bounding_box_format="xyxy", class_mapping=class_mapping)

内容的提问来源于stack exchange，提问作者Marlies Lauwers