Question

正如标题所暗示的，预测的输出为1.0或0.0，并且我没有得到任何介于两者之间的值，这导致我的输出下降了100％或下降了100％。

for img in imagesList:
    test_image = image.load_img(path+img, target_size=(64, 64))
    test_images = image.img_to_array(test_image)
    test_images = np.expand_dims(test_image, axis = 0)
    result = classifier.predict(test_images)
    if result[0][0] <= 0.0:
      prediction = 'fall'

      print('The file is:{} while model output is: {}{} {} '.format(img[:-4],((1-result[0][0])*100),'%',prediction))
      print(result[0][0])
      #print('The file is:{} while model output is: {}{} {}'.format(img[:-4],((result[0][0])*100),'%',' nonfall'))
    else:
      prediction = 'nonfall'
      print('The file is:{} while model output is: {}{} {} '.format(img[:-4],((result[0][0])*100),'%',prediction))
      print(result[0][0])

Answer 1

在每个import numpy as np import matplotlib.pyplot as plt x = np.arange(-5, 5, 0.01) # need higher resolution to get nice clipping boundaries y = np.arange(-5, 5, 0.01) xx, yy = np.meshgrid(x, y) # need dense meshgrid for contains_points z = np.sin(xx**2 + yy**2) / (xx**2 + yy**2) patch = plt.Circle((-2, 2), 1) # circle with radius 1 around the point (-2, 2) path = patch.get_path().transformed(patch.get_patch_transform()) # the path attribute of Circle has to be transformed to data units is_valid = path.contains_points(np.c_[xx.ravel(), yy.ravel()]).reshape(xx.shape) z[~is_valid] = np.nan # NaN values are not plotted h = plt.contourf(xx,yy,z) plt.plot(-2, 2.5, 'or') plt.show()循环迭代中，您将获得单个图像，并且该图像的预测要么是类别，要么是类别。如果要获得批次的准确性，则需要附加批次的图像，然后进行预测。代码应该看起来像这样。

for

Answer 2

编辑：

我通过使用model.predict_prob(img)解决了这个问题。

引用：https://stackoverflow.com/a/54532343/13467194

CNN模型的输出为1.0或0.0

2 个答案: