我试图NasNetlarge在没有顶部的 Keras 中使用，但我无法摆脱顶部：
from keras.applications import NASNetLarge
model = NASNetLarge(input_shape=(224, 224, 3), include_top=False,
weights="imagenet")
如果我使用 (331, 331, 3) 形状以外的任何形状，include_top=True即使我设置了include_top=False. 如果我将输入形状设置为 (331, 331, 3) 和include_top，False它会下载顶部的权重。我做错了什么还是 Keras 模块中有错误？
你的keras版本是什么？您可以尝试的两件事是：
如果以上不行，在keras中打开nasnet.py文件，修改这部分代码：
elif default_size == 331: # large version
if include_top:
weight_path = NASNET_LARGE_WEIGHT_PATH
model_name = 'nasnet_large.h5'
else:
weight_path = NASNET_LARGE_WEIGHT_PATH_NO_TOP
model_name = 'nasnet_large_no_top.h5'
weights_file = get_file(model_name, weight_path,
cache_subdir='models')
model.load_weights(weights_file)
对此：
elif default_size == 331: # large version
if include_top:
weight_path = NASNET_LARGE_WEIGHT_PATH
model_name = 'nasnet_large.h5'
else:
weight_path = NASNET_LARGE_WEIGHT_PATH_NO_TOP
model_name = 'nasnet_large_no_top.h5'
weights_file = get_file(model_name, weight_path,
cache_subdir='models')
model.load_weights(weights_file, by_name = True)