如何训练mobilenet——v2模型

[KelvinHuang666]

修改代码直接调用会有bug
修改如下：
with tf.contrib.slim.arg_scope(mobilenet_v2.training_scope()):
out, endpoints = mobilenet_v2.mobilenet(input_tensor=input_images)

BUG如下：
Traceback (most recent call last):
File "train_v2.py", line 213, in
snapshot_prefix=snapshot_prefix)
File "train_v2.py", line 158, in train
tf.losses.softmax_cross_entropy(onehot_labels=input_labels, logits=out)
File "/opt/conda/lib/python3.6/site-packages/tensorflow/python/ops/losses/losses_impl.py", line 732, in softmax_cross_entropy
logits.get_shape().assert_is_compatible_with(onehot_labels.get_shape())
File "/opt/conda/lib/python3.6/site-packages/tensorflow/python/framework/tensor_shape.py", line 764, in assert_is_compatible_with
raise ValueError("Shapes %s and %s are incompatible" % (self, other))
ValueError: Shapes (?, 1001) and (?, 4) are incompatible

[Ijustakid]

mobilenet_v2.py的labels_num=1001，看你的分类多少把1001改掉