TensorFlow中YOLOv3模型报错：'list' object has no attribute 'shape'

我使用TensorFlow编写代码时，尝试打印模型输出的shape，出现了'list' object has no attribute 'shape'的报错。

相关代码 #

推理代码 #

image_batch, label_batch = next(iter(train_ds))
feature_batch = train_model(image_batch)
print(feature_batch.shape)

YOLOv3模型构建与保存代码 #

# define the model
model = make_yolov3_model()

# load the model weights
# I have loaded the pretrained weights in a separate dataset
weight_reader = WeightReader('yolov3.weights')

# set the model weights into the model
weight_reader.load_weights(model)

# save the model to file
model.save('model.h5')

模型加载代码 #

# load yolov3 model
from keras.models import load_model
train_model = load_model('model.h5', compile=False)

报错信息 #

AttributeError                            Traceback (most recent call last)
~\AppData\Local\Temp\ipykernel_4412\2773113981.py in <module>
1 image_batch, label_batch = next(iter(train_ds))
2 feature_batch = train_model(image_batch)
----> 3 print(feature_batch.shape)

AttributeError: 'list' object has no attribute 'shape'

原因与解决办法 #

原因 #

YOLOv3是多输出模型，它会生成3个不同尺度的检测输出（对应不同尺寸的锚框，用于检测大小不同的目标），因此调用train_model(image_batch)返回的不是单个张量，而是包含3个张量的列表，列表没有shape属性，所以触发报错。

解决办法 #

• 查看所有输出的shape：遍历列表逐个打印

feature_batch = train_model(image_batch)
for idx, output_tensor in enumerate(feature_batch):
    print(f"第{idx+1}个输出的shape: {output_tensor.shape}")

• 查看单个输出的shape：通过索引直接取对应位置的张量后打印

feature_batch = train_model(image_batch)
# 示例：取第一个输出的shape
print(feature_batch[0].shape)

内容的提问来源于stack exchange，提问作者Muhammad Rafi