faster-rcnn错误信息 ： tensorflow.python.framework.errors_impl.InvalidArgumentError: Assign requires shapes of both tensors to match. lhs shape= [21] rhs shape= [2]
1 Traceback (most recent call last): 2 File "C:UsersAdministratorAppDataLocalProgramsPythonPython36libsite-packages ensorflowpython rainingsaver.py", line 1276, in restore 3 {self.saver_def.filename_tensor_name: save_path}) 4 File "C:UsersAdministratorAppDataLocalProgramsPythonPython36libsite-packages ensorflowpythonclientsession.py", line 929, in run 5 run_metadata_ptr) 6 File "C:UsersAdministratorAppDataLocalProgramsPythonPython36libsite-packages ensorflowpythonclientsession.py", line 1152, in _run 7 feed_dict_tensor, options, run_metadata) 8 File "C:UsersAdministratorAppDataLocalProgramsPythonPython36libsite-packages ensorflowpythonclientsession.py", line 1328, in _do_run 9 run_metadata) 10 File "C:UsersAdministratorAppDataLocalProgramsPythonPython36libsite-packages ensorflowpythonclientsession.py", line 1348, in _do_call 11 raise type(e)(node_def, op, message) 12 tensorflow.python.framework.errors_impl.InvalidArgumentError: Assign requires shapes of both tensors to match. lhs shape= [21] rhs shape= [2] 13 [[node save/Assign_517 (defined at G:/jianfeng/project/rubblish_det/faster_rcnn/tf-faster-rcnn-master/tools/demo.py:149) ]]
我的代码来自 https://github.com/endernewton/tf-faster-rcnn
这是由于类别数目不一致，导致的 ；看错误信息倒数第二行，lhs为21类 rhs为2类
解决方法：在demo.py文件中找到下面两行代码，将 "TEST"后的21改为 2 因为我的数据类别是 background+单类别=2
1 net.create_architecture("TEST", 21, 2 tag='default', anchor_scales=[8, 16, 32])