NotImplementedError: Cannot convert a symbolic Tensor (truediv:0) to a numpy array.

Question

用训练好的模型识别自己手写的图片时报错
def load_preprosess_image(path):
img_raw = tf.io.read_file(path)
img_tensor = tf.cast(tf.image.decode_jpeg(img_raw, channels=1), tf.float32)
img_tensor = tf.image.resize(img_tensor, [28, 28])
img = img_tensor / 255.0
return img

if name == 'main':
app = Painter('./img/')
app.run()
ID = []
for count in range(app.count):
img = load_preprosess_image('./img/' + str(count) + '.jpg')

    np.expand_dims(img, axis=-1)

    model = tf.keras.models.load_model('./model.h5')
    prediction = model.predict(img.numpy().reshape(1, 28, 28, 1))
    result = np.argmax(prediction)
    ID.append(result)
print('识别的数字为：', end='')
for item in ID:
    print(item, end='')

环境Python3.6，tensorflow1.15，看到网上说是numpy版本的问题，我卸载分别重新装了1.17，1.18，1.19版本，还是报错，

Answer 1

不要用tf.cast，用np的函数

解决Cannot convert a symbolic Tensor to a numpy array._yulanf的博客-CSDN博客_symbolic tensor 今天做Tensorflow的手写体识别模型，用的Lenet5,跑完之后想自己输入个图片测试一下，处理图片并导入数据的时候一直报错，Cannot convert a symbolic Tensor to a numpy array.最后发现reshaped_xs = np.reshape(xs_c, (BATCH_SIZE, mnist_inference_Lenet5_update.IMAGE... https://blog.csdn.net/yulanf/article/details/104494492