我想像https://www.tensorflow.org/guide/keras/train_and_evaluate#specifying_a_loss_metrics_and_an_optimizer
中那样自定义指标我的代码如下
class IOU(tf.keras.metrics.Metric):
def __init__(self, name='iou_part', **kwargs):
super(IOU, self).__init__(name=name, **kwargs)
self.iou = self.add_weight(name='iou_part', initializer='zeros')
self.template_width = 115
self.template_height = 75
self.frame_width = 1280
self.frame_height = 720
self.corners = tf.constant([[-0.5, 0.1], [-0.5, 0.5], [0.5, 0.5], [0.5, 0.1]], dtype=tf.float32)
self.epsilon = 1e-6
def update_state(self, y_true, y_pred, sample_weight=None):
batch_size = y_true.shape[0]
fake_frame = tf.ones((batch_size, 1, self.frame_height, self.frame_width))
fake_template = tf.ones((batch_size, 1, self.template_height, self.template_width))
target = get_perspective_transform(self.corners, tf.reshape(y_true, (-1, 2, 4)))
output = get_perspective_transform(self.corners, tf.reshape(y_pred, (-1, 2, 4)))
## Compute IOU
但是，这会产生错误“ TypeError：预期为int32，而没有类型为“ NoneType”。这是因为执行model.compile（....）时y_true为（None，4，2）。将批次大小纳入指标的正确方法是什么？