如何找到用于tensorflow对象检测的序列在keras中的输入参数

Question

我成功地在google colab上运行TensorFlow Hub Object Detection Colab的推理

我加载的模型

model_display_name = 'CenterNet HourGlass104 Keypoints 512x512'
model_handle = 'https://tfhub.dev/tensorflow/centernet/hourglass_512x512/1'

尝试在TensorFlow Hub中使用迁移学习

IMAGE_SHAPE = (None, None)
classifier = tf.keras.Sequential([
hub.KerasLayer(model_handle, input_shape=IMAGE_SHAPE+(3,))
])

我得到这个错误

ValueError                                Traceback (most recent call last)
<ipython-input-23-081693dbe40a> in <module>()
4 
5 object_detector = tf.keras.Sequential([
----> 6     hub.KerasLayer(object_detector_model, input_shape=Object_Detector_IMAGE_SHAPE+(3,))
7 ])
2 frames
/usr/local/lib/python3.7/dist-packages/tensorflow/python/autograph/impl/api.py in wrapper(*args, **kwargs)
690       except Exception as e:  # pylint:disable=broad-except
691         if hasattr(e, 'ag_error_metadata'):
--> 692           raise e.ag_error_metadata.to_exception(e)
693         else:
694           raise
ValueError: Exception encountered when calling layer "keras_layer" (type KerasLayer).
in user code:
File "/usr/local/lib/python3.7/dist-packages/tensorflow_hub/keras_layer.py", line 229, in call  *
result = f()
ValueError: Python inputs incompatible with input_signature:
inputs: (
Tensor("Placeholder:0", shape=(None, 1, None, None, 3), dtype=float32))
input_signature: (
TensorSpec(shape=(1, None, None, 3), dtype=tf.uint8, name=None)).

Call arguments received:
• inputs=tf.Tensor(shape=(None, 1, None, None, 3), dtype=float32)
• training=None

我如何改变IMAGE_SHAPE为此，我很困惑?

需要帮助，谢谢你

Answer 1

有两个问题，该模型在指定input_shape时期望输入类型为tf.uint8，并根据源返回输出字典。由于SequentialAPI不能与多个输出一起工作，您将不得不使用FunctionalAPI。

简单的例子:

import tensorflow as tf
import tensorflow_hub as hub
model_display_name = 'CenterNet HourGlass104 Keypoints 512x512'
model_handle = 'https://tfhub.dev/tensorflow/centernet/hourglass_512x512/1'
k_layer = hub.KerasLayer(model_handle, input_shape=(None, None, 3), dtype=tf.uint8)

和FunctionalAPI:

import tensorflow as tf
import tensorflow_hub as hub
model_display_name = 'CenterNet HourGlass104 Keypoints 512x512'
model_handle = 'https://tfhub.dev/tensorflow/centernet/hourglass_512x512/1'
k_layer = hub.KerasLayer(model_handle, input_shape=(None, None, 3))
inputs = tf.keras.layers.Input(shape=(None, None, 3), dtype=tf.uint8)
outputs = k_layer(inputs)
classifier = tf.keras.Model(inputs, outputs)
print(classifier.summary())

Model: "model"
_________________________________________________________________
Layer (type)                Output Shape              Param #   
=================================================================
input_2 (InputLayer)        [(None, None, None, 3)]   0         

keras_layer_9 (KerasLayer)  {'num_detections': (1,),  0         
'detection_boxes': (1,             
100, 4),                            
'detection_classes': (1            
, 100),                             
'detection_scores': (1,            
100)}                              

=================================================================
Total params: 0
Trainable params: 0
Non-trainable params: 0
_________________________________________________________________
None