I created the TFRecord data and then train the model with coco-stuff dataset，then error occured in evaluate.py

[HanChen-HUST]

tensorflow.python.framework.errors_impl.InvalidArgumentError: Number of ways to split should evenly divide the split dimension, but got split_dim 2(size=1) and num_split 3 [[{{node split}} = Split[T=DT_FLOAT, num_split=3, _device="/device:CPU:0"](split/split_dim, ToFloat)]] [{{node IteratorGetNext}} = IteratorGetNextoutput_shapes=[[?,?,?,3],[?,?,?,1]], output_types=[DT_FLOAT，DT_INT32], _device="/job:localhost/replica:0/task:0/device:CPU:0"]] Hope to get your answer,that will be great apperciate

[ytliang97]

@chenhust1995 #38 maybe could help you.