张量流转换keras模型的tensorflowjs

时间:2018-04-27 23:46:06

标签: python tensorflow tensorflow.js

使用标准代码     导入张量流为tf      将tensorflowjs导入为tfjs

 tfjs_target_dir = 'c:/files'
 model = tf.keras.applications.InceptionV3(
     include_top=True,
     weights='imagenet',
     input_tensor=None,
     input_shape=None,
     pooling=None,
     classes=1000
 )
 tfjs.converters.save_keras_model(model, tfjs_target_dir)

我收到以下错误

文件" C:\ Program Files \ python36 \ lib \ site-packages \ tensorflow \ python \ client \ session.py",第1321行,在_do_run中     run_metadata)

文件" C:\ Program Files \ python36 \ lib \ site-packages \ tensorflow \ python \ client \ session.py",第1340行,在_do_call     提升类型(e)(node_def,op,message)

FailedPreconditionError:尝试使用未初始化的值conv2d_98 / kernel      [[Node:_retval_conv2d_98 / kernel_0_0 = _RetvalT = DT_FLOAT,index = 0,_device =" / job:localhost / replica:0 / task:0 / device:CPU:0"]]

1 个答案:

答案 0 :(得分:0)

您是否尝试过提供输入形状?

e.g。

if K.image_data_format() == 'channels_first':
    input_shape = (3, img_height, img_width)
else:
    input_shape = (img_height, img_width, 3)

img_height / width是您要预测的图片的大小