Python 在Keras的本地化

我有一个机器学习CNN，它应该对图像中的物体进行分类和定位。输入是一个尺寸为（448、448、3）的RGB图像，输出是24（对于边界框坐标，因为图像中有多个对象，最多6个，图像的id为6）。下面是我如何定义模型的：

def define_model():

    input = Input(shape = (448,448,3)
    x = Conv2D(32, (3,3), activation = 'relu')(input)
    x = MaxPooling2D((3,3))(x)
    x = Conv2D(64, (3,3), activation = 'relu')(x)
    x = MaxPooling2D((3,3))(x)
    x = Conv2D(64, (3,3), activation = 'relu')(x)
    x = Conv2D(32, (3,3), activation = 'relu')(x)
    x = GlobalAveragePooling2D()(x)
    
    classification_output_head = Flatten()(x)
    classification_output_head = Dropout(0.1)(classification_output_head)
    classification_output_head = Dense(6, activation = 'sigmoid', name = 'class_of_objs')(classification_output_head)
    
    regression_for_bounding_box_output_head = Flatten()(x)
    regression_for_bounding_box_output_head = Dense(64, activation = 'relu')((regression_for_bounding_box_output_head))
    regression_for_bounding_box_output_head = Dense(32, activation = 'relu')(regression_for_bounding_box_output_head)
    regression_for_bounding_box_output_head = Dense(24, activation =  'sigmoid', name = 'bounding_box')(regression_for_bounding_box_output_head)

return Model(inputs=[input], outputs=[classification_output_head, regression_for_bounding_box_output_head])

model = define_model()

然后我定义模型：

def define_model():

    input = Input(shape = (448,448,3)
    x = Conv2D(32, (3,3), activation = 'relu')(input)
    x = MaxPooling2D((3,3))(x)
    x = Conv2D(64, (3,3), activation = 'relu')(x)
    x = MaxPooling2D((3,3))(x)
    x = Conv2D(64, (3,3), activation = 'relu')(x)
    x = Conv2D(32, (3,3), activation = 'relu')(x)
    x = GlobalAveragePooling2D()(x)
    
    classification_output_head = Flatten()(x)
    classification_output_head = Dropout(0.1)(classification_output_head)
    classification_output_head = Dense(6, activation = 'sigmoid', name = 'class_of_objs')(classification_output_head)
    
    regression_for_bounding_box_output_head = Flatten()(x)
    regression_for_bounding_box_output_head = Dense(64, activation = 'relu')((regression_for_bounding_box_output_head))
    regression_for_bounding_box_output_head = Dense(32, activation = 'relu')(regression_for_bounding_box_output_head)
    regression_for_bounding_box_output_head = Dense(24, activation =  'sigmoid', name = 'bounding_box')(regression_for_bounding_box_output_head)

return Model(inputs=[input], outputs=[classification_output_head, regression_for_bounding_box_output_head])

model = define_model()

以下是剩下的：

losses = {'class_of_objs': 'sparse_categorical_crossentropy',
          'bounding_box': 'mse'}
model.compile('adam', loss=losses, metrics=['mae'])
tloader = tf.data.Dataset.from_tensor_slices((x, y_class, y_bbox))
model.fit(tloader,epochs=100)

运行model.fit时，出现以下错误：

Epoch 1/100

WARNING:tensorflow:Model was constructed with shape (None, 448, 448, 3) for input Tensor("input_25:0", shape=(None, 448, 448, 3), dtype=float32), but it was called on an input with incompatible shape (448, 448, 3).
--------------------------------------------------------------------------- ValueError                                Traceback (most recent call last) <ipython-input-308-70dc0cc6919c> in <module>
----> 1 model.fit(tloader,epochs=100)

~/opt/anaconda3/lib/python3.8/site-packages/tensorflow/python/keras/engine/training.py in _method_wrapper(self, *args, **kwargs)
    106   def _method_wrapper(self, *args, **kwargs):
    107     if not self._in_multi_worker_mode():  # pylint: disable=protected-access
--> 108       return method(self, *args, **kwargs)
    109 
    110     # Running inside `run_distribute_coordinator` already.

~/opt/anaconda3/lib/python3.8/site-packages/tensorflow/python/keras/engine/training.py in fit(self, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, validation_batch_size, validation_freq, max_queue_size, workers, use_multiprocessing)    1096                 batch_size=batch_size):   1097               callbacks.on_train_batch_begin(step)
-> 1098               tmp_logs = train_function(iterator)    1099               if data_handler.should_sync:    1100                 context.async_wait()

~/opt/anaconda3/lib/python3.8/site-packages/tensorflow/python/eager/def_function.py in __call__(self, *args, **kwds)
    778       else:
    779         compiler = "nonXla"
--> 780         result = self._call(*args, **kwds)
    781 
    782       new_tracing_count = self._get_tracing_count()

~/opt/anaconda3/lib/python3.8/site-packages/tensorflow/python/eager/def_function.py in _call(self, *args, **kwds)
    821       # This is the first call of __call__, so we have to initialize.
    822       initializers = []
--> 823       self._initialize(args, kwds, add_initializers_to=initializers)
    824     finally:
    825       # At this point we know that the initialization is complete (or less

~/opt/anaconda3/lib/python3.8/site-packages/tensorflow/python/eager/def_function.py in _initialize(self, args, kwds, add_initializers_to)
    694     self._graph_deleter = FunctionDeleter(self._lifted_initializer_graph)
    695     self._concrete_stateful_fn = (
--> 696         self._stateful_fn._get_concrete_function_internal_garbage_collected( 
# pylint: disable=protected-access
    697             *args, **kwds))
    698 

~/opt/anaconda3/lib/python3.8/site-packages/tensorflow/python/eager/function.py in _get_concrete_function_internal_garbage_collected(self, *args,
**kwargs)    2853       args, kwargs = None, None    2854     with self._lock:
-> 2855       graph_function, _, _ = self._maybe_define_function(args, kwargs)    2856     return graph_function    2857 

~/opt/anaconda3/lib/python3.8/site-packages/tensorflow/python/eager/function.py in _maybe_define_function(self, args, kwargs)    3211     3212       self._function_cache.missed.add(call_context_key)
-> 3213       graph_function = self._create_graph_function(args, kwargs)    3214       self._function_cache.primary[cache_key] = graph_function    3215       return graph_function, args, kwargs

~/opt/anaconda3/lib/python3.8/site-packages/tensorflow/python/eager/function.py in _create_graph_function(self, args, kwargs, override_flat_arg_shapes)    3063     arg_names = base_arg_names + missing_arg_names    3064     graph_function = ConcreteFunction(
-> 3065         func_graph_module.func_graph_from_py_func(    3066             self._name,    3067             self._python_function,

~/opt/anaconda3/lib/python3.8/site-packages/tensorflow/python/framework/func_graph.py in func_graph_from_py_func(name, python_func, args, kwargs, signature, func_graph, autograph, autograph_options, add_control_dependencies, arg_names, op_return_value, collections, capture_by_value, override_flat_arg_shapes)
    984         _, original_func = tf_decorator.unwrap(python_func)
    985 
--> 986       func_outputs = python_func(*func_args, **func_kwargs)
    987 
    988       # invariant: `func_outputs` contains only Tensors, CompositeTensors,

~/opt/anaconda3/lib/python3.8/site-packages/tensorflow/python/eager/def_function.py in wrapped_fn(*args, **kwds)
    598         # __wrapped__ allows AutoGraph to swap in a converted function. We give
    599         # the function a weak reference to itself to avoid a reference cycle.
--> 600         return weak_wrapped_fn().__wrapped__(*args, **kwds)
    601     weak_wrapped_fn = weakref.ref(wrapped_fn)
    602 

~/opt/anaconda3/lib/python3.8/site-packages/tensorflow/python/framework/func_graph.py in wrapper(*args, **kwargs)
    971           except Exception as e:  # pylint:disable=broad-except
    972             if hasattr(e, "ag_error_metadata"):
--> 973               raise e.ag_error_metadata.to_exception(e)
    974             else:
    975               raise

纪元1/100
警告：tensorflow:为输入张量（“input_25:0”，shape=（None，448，448，3），dtype=float32）构造了形状（None，448，448，3）的模型，但在具有不兼容形状（448，448，3）的输入上调用了该模型。
---------------------------------------------------------------------------中的ValueError回溯（最近一次调用）
---->1型号配合（t装载机，历代=100）
~/opt/anaconda3/lib/python3.8/site-packages/tensorflow/python/keras/engine/training.py in\u method\u包装（self，*args，**kwargs）
106定义方法包装（self，*args，**kwargs）：
107如果不是self._处于_multi_worker_模式（）：#pylint:disable=受保护的访问
-->108返回方法（self、*args、**kwargs）
109
110#已经在“运行分配协调器”内部运行了。
~/opt/anaconda3/lib/python3.8/site-packages/tensorflow/python/keras/engine/training.py in-fit（self、x、y、批大小、历元、冗余、回调、验证拆分、验证数据、洗牌、类权重、样本权重、初始历元、每个历元的步骤、验证步骤、验证批次大小、验证频率、最大队列大小、工作人员、使用多处理）1096批大小=批大小）：1097回调。列车上的批处理开始（步骤）
->1098 tmp_logs=训练函数（迭代器）1099 if data_handler.should_sync:1100 context.async_wait（）
~/opt/anaconda3/lib/python3.8/site-packages/tensorflow/python/eager/def_function.py在调用中（self，*args，**kwds）
778其他：
779 compiler=“nonXla”
-->780结果=自身调用（*args，**kwds）
781
782 new_tracing_count=self._get_tracing_count（）
~/opt/anaconda3/lib/python3.8/site-packages/tensorflow/python/eager/def_function.py in_调用（self，*args，**kwds）
821#这是"调用"的第一个调用，因此我们必须初始化。
822个初始值设定项=[]
-->823自身初始化（参数、KWD、添加初始化器到=初始化器）
824最后：
825#此时我们知道初始化已完成（或更少）
~/opt/anaconda3/lib/python3.8/site-packages/tensorflow/python/eager/def_function.py in_initialize（self、args、kwds、add_initializers_to）
694 self.\u graph\u deleter=函数deleter（self.\u lifted\u initializer\u graph）
695自具体状态fn=(
-->696自我状态功能获取具体功能内部垃圾收集（
#pylint:disable=受保护的访问
697*args，**科威特第纳尔）
698
~/opt/anaconda3/lib/python3.8/site-packages/tensorflow/python/eager/function.py in\u get\u concrete\u function\u internal\u garbage\u collected（self，*args，
**kwargs）2853 args，kwargs=None，None 2854带self.\u锁：
->2855图形函数，可能定义函数（args，kwargs）2856返回图形函数2857
~/opt/anaconda3/lib/python3.8/site-packages/tensorflow/python/eager/function.py in\u maybe\u define\u function（self、args、kwargs）3211 3212 self.\u function\u cache.missed.add（调用上下文键）
->3213图形函数=自。\创建图形函数（args，kwargs）3214自。\函数\u缓存。主[缓存\u键]=图形函数3215返回图形函数，args，kwargs
~/opt/anaconda3/lib/python3.8/site-packages/tensorflow/python/eager/function.py in\u create\u graph\u function（self、args、kwargs、override\u flat\u arg\u shapes）3063 arg\u names=基本arg\u names+缺少arg\u names 3064 graph\u function=具体函数(
->3065函数图模块。函数图来自于函数（3066 self.\u名称，3067 self.\u python函数，
~/opt/anaconda3/lib/python3.8/site-packages/tensorflow/python/framework/func\u graph.py in func\u graph\u from\u py func（name，python\u func，args，kwargs，signature，func\u graph，autograph，options，add\u control\u依赖项，arg\u名称，op\u返回值，集合，按值捕获，覆盖平面arg\u形状）
984，original_func=tf_decorator.unwrap（python_func）
985
-->986 func_输出=python_func（*func_参数，**func_参数）
987
988#不变量：`func_outputs`只包含张量、复合传感器、，
~/opt/anaconda3/lib/python3.8/site-packages/tensorflow/python/eager/def_function.py in wrapped_fn（*args，**kwds）
598#uuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu
599#函数对自身进行弱引用以避免引用循环。
-->600返回弱_-wrapped_-fn（）
601弱包层=弱包层参考（包层）
602
包装器中的~/opt/anaconda3/lib/python3.8/site-packages/tensorflow/python/framework/func_graph.py（*args，**kwargs）
971例外情况为e:#pylint:disable=broad except
972如果hasattr（e，“AGU错误元数据”）：
-->973将e.ag\u错误\u元数据引发到\u异常（e）
974其他：
975提高

---

如何修复它在调用model.fit之前，您必须使用model.complile编译模型，在该模型上设置losse，并按照以下方式选择优化器：

model.compile（optimizer=“Adam”，loss=“mse”，metrics=[“mae”]）

很抱歉，我确实这么做了，但没有把它放在帖子里。您的回溯海报不完整。@Dr.Snoopy您能详细说明一下吗，我似乎不明白您的意思