tensorflow - 从文件加载权重时层数不匹配。ImageAi 自定义

Question

我需要帮助...我使用 imageai 自定义类来创建我自己的检测...

现在我们开始

from imageai.Classification.Custom import ClassificationModelTrainer
model_trainer = ClassificationModelTrainer()
model_trainer.setModelTypeAsResNet50()
model_trainer.setDataDirectory("leads_test")

model_trainer.trainModel(num_objects=1, num_experiments=1, enhance_data=True, batch_size=1, show_network_summary=True)

<...>

from imageai.Detection import ObjectDetection

detector = ObjectDetection()
model_path = "leads_test/models/model_ex-001_acc-1.000000.h5"
input_path = "ECG/IMG_0239.jpg"
output_path = "./output/newimage.jpg"

detector.setModelTypeAsTinyYOLOv3()
detector.setModelPath(model_path)
detector.loadModel()

ValueError: Layer count mismatch when loading weights from file. Model expected 24 layers, found 107 saved layers.

Answer 1

解决了。我必须imageai.Classification.Custom import CustomImageClassification改用of imageai.Detection import ObjectDetection

from imageai.Classification.Custom import CustomImageClassification
prediction = CustomImageClassification()
prediction.setModelTypeAsResNet50()
prediction.setModelPath('leads_test/models/model_ex-001_acc-1.000000.h5')
prediction.setJsonPath('leads_test/json/model_class.json')
prediction.loadModel(num_objects=1)

问题在于训练和预测时的模型类型不同。