之前在tensorflow2.0版本使用以下搭配可以成功
class MeanIoU(tf.keras.metrics.MeanIoU):
"""MeanIoU for sparse_categorical_crossentropy"""
def __call__(self, y_true, y_pred, sample_weight=None):
y_pred = tf.argmax(y_pred, axis=-1)
return super().__call__(y_true, y_pred, sample_weight=sample_weight)
model.compile(optimizer=optimizer, loss='sparse_categorical_crossentropy',
metrics=['accuracy', MeanIoU(num_classes=CLASSES, name='mIOU')])
但是升级到了Tensorflow2.3.0版本一直出现以下维度的问题
Shapes of all inputs must match: values[0].shape = [80] != values[1].shape = [400] （num_class=5）
更新后可以正常使用了。
class UpdatedMeanIoU(tf.keras.metrics.MeanIoU):
def __init__(self,
y_true=None,
y_pred=None,
num_classes=None,
name=None,
dtype=None):
super(UpdatedMeanIoU, self).__init__(num_classes = num_classes,name=name, dtype=dtype)
def update_state(self, y_true, y_pred, sample_weight=None):
y_pred = tf.math.argmax(y_pred, axis=-1)
return super().update_state(y_true, y_pred, sample_weight)
model.compile(optimizer=optimizer, loss='sparse_categorical_crossentropy',
metrics=['accuracy', UpdatedMeanIoU(num_classes=CLASSES, name='mIOU')])
