yyuananyvision/mxnet_center_loss

implement center loss operator for mxnet

mxnet_center_loss's Issues

About center update

I think the scale parameter should be multiplied duiring updating center.
Since the scale is a constant to weight center loss, it should influence both data sample and center backward.

请求更新支援 mxnet 0.10.1

请求更新到 mxnet 0.10.1的版本

AttributeError: 'Accuracy' object has no attribute 'num'

hi,can you help me address the problem?
run train.py
come across this wrong:AttributeError: 'Accuracy' object has no attribute 'num'

无法执行此项目

想测试一下center loss 的威力...
git clone 下来 ,到 mxnet的项目中把 test/python/common/get_data.py 复制到此项目下,
执行train.py
执行失败

系统 Ubuntu 16.04 x64
CUDA 8.0
Cudnn 6.0
mxnet 0.10.1 and 0.11.0
opencv 3.3

求各路大神帮帮忙
@pangyupo @luoyetx

`[14:48:02] src/io/iter_mnist.cc:94: MNISTIter: load 60000 images, shuffle=1, shape=(100,1,28,28)
[14:48:02] src/io/iter_mnist.cc:94: MNISTIter: load 10000 images, shuffle=1, shape=(100,1,28,28)
training model ...
dev is [gpu(0)]
/home/mxnet_center_loss/train_model.py:133: DeprecationWarning: mxnet.model.FeedForward has been deprecated. Please use mxnet.mod.Module instead.
**model_args)
/home/mxnet/python/mxnet/initializer.py:353: DeprecationWarning: Calling initializer with init(str, NDArray) has been deprecated.please use init(mx.init.InitDesc(...), NDArray) instead.
init(name, arr)
[14:48:12] /home/mxnet/dmlc-core/include/dmlc/./logging.h:308: [14:48:12] src/pass/gradient.cc:159: Check failed: (*rit)->inputs.size() == input_grads.size() (5 vs. 2) Gradient function not returning enough gradient

Stack trace returned 10 entries:
[bt] (0) /home/mxnet/python/mxnet/../../lib/libmxnet.so(_ZN4dmlc15LogMessageFatalD1Ev+0x3c) [0x7f28b1f644bc]
[bt] (1) /home/mxnet/python/mxnet/../../lib/libmxnet.so(+0x27b1f40) [0x7f28b4021f40]
[bt] (2) /home/mxnet/python/mxnet/../../lib/libmxnet.so(ZNSt17_Function_handlerIFN4nnvm5GraphES1_EPS2_E9_M_invokeERKSt9_Any_dataOS1+0x111) [0x7f28b2c033f1]
[bt] (3) /home/mxnet/python/mxnet/../../lib/libmxnet.so(_ZN4nnvm11ApplyPassesENS_5GraphERKSt6vectorINSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEESaIS7_EE+0x32c) [0x7f28b4053b8c]
[bt] (4) /home/mxnet/python/mxnet/../../lib/libmxnet.so(_ZN4nnvm9ApplyPassENS_5GraphERKNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEE+0x3c9) [0x7f28b2f99389]
[bt] (5) /home/mxnet/python/mxnet/../../lib/libmxnet.so(_ZN4nnvm4pass8GradientENS_5GraphESt6vectorINS_9NodeEntryESaIS3_EES5_S5_St8functionIFS3_OS5_EES6_IFiRKNS_4NodeEEES6_IFS3_RKS3_SG_EES2_IPKNS_2OpESaISL_EENSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEE+0x60c) [0x7f28b300780c]
[bt] (6) /home/mxnet/python/mxnet/../../lib/libmxnet.so(_ZN5mxnet4exec13GraphExecutor13InitFullGraphEN4nnvm6SymbolERKSt6vectorINS_9OpReqTypeESaIS5_EE+0x863) [0x7f28b2ff1283]
[bt] (7) /home/mxnet/python/mxnet/../../lib/libmxnet.so(_ZN5mxnet4exec13GraphExecutor9InitGraphEN4nnvm6SymbolERKNS_7ContextERKSt3mapINSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEES4_St4lessISD_ESaISt4pairIKSD_S4_EEERKSt6vectorIS4_SaIS4_EESR_SR_RKSN_INS_9OpReqTypeESaISS_EE+0x82) [0x7f28b2ff1b52]
[bt] (8) /home/mxnet/python/mxnet/../../lib/libmxnet.so(_ZN5mxnet4exec13GraphExecutor4InitEN4nnvm6SymbolERKNS_7ContextERKSt3mapINSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEES4_St4lessISD_ESaISt4pairIKSD_S4_EEERKSt6vectorINS_7NDArrayESaISO_EESS_RKSN_INS_9OpReqTypeESaIST_EESS_PNS_8ExecutorERKSt13unordered_mapINS2_9NodeEntryESO_NS2_13NodeEntryHashENS2_14NodeEntryEqualESaISG_IKS11_SO_EEE+0x76f) [0x7f28b2ffc45f]
[bt] (9) /home/mxnet/python/mxnet/../../lib/libmxnet.so(ZN5mxnet8Executor4BindEN4nnvm6SymbolERKNS_7ContextERKSt3mapINSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEES3_St4lessISC_ESaISt4pairIKSC_S3_EEERKSt6vectorINS_7NDArrayESaISN_EESR_RKSM_INS_9OpReqTypeESaISS_EESR_PS0+0x1b8) [0x7f28b2ffd7f8]

Traceback (most recent call last):
File "./train.py", line 97, in
main()
File "./train.py", line 94, in main
train_model.fit(args, net, (train, val), data_shape)
File "/home/mxnet_center_loss/train_model.py", line 153, in fit
epoch_end_callback = checkpoint)
File "/home/mxnet/python/mxnet/model.py", line 830, in fit
sym_gen=self.sym_gen)
File "/home/mxnet/python/mxnet/model.py", line 210, in _train_multi_device
logger=logger)
File "/home/mxnet/python/mxnet/executor_manager.py", line 326, in init
self.slices, train_data)
File "/home/mxnet/python/mxnet/executor_manager.py", line 238, in init
input_types=data_types)
File "/home/mxnet/python/mxnet/executor_manager.py", line 184, in _bind_exec
grad_req=grad_req, shared_exec=base_exec)
File "/home/mxnet/python/mxnet/symbol.py", line 1636, in bind
ctypes.byref(handle)))
File "/home/mxnet/python/mxnet/base.py", line 102, in check_call
raise MXNetError(py_str(_LIB.MXGetLastError()))
mxnet.base.MXNetError: [14:48:12] src/pass/gradient.cc:159: Check failed: (*rit)->inputs.size() == input_grads.size() (5 vs. 2) Gradient function not returning enough gradient

Stack trace returned 10 entries:
[bt] (0) /home/mxnet/python/mxnet/../../lib/libmxnet.so(_ZN4dmlc15LogMessageFatalD1Ev+0x3c) [0x7f28b1f644bc]
[bt] (1) /home/mxnet/python/mxnet/../../lib/libmxnet.so(+0x27b1f40) [0x7f28b4021f40]
[bt] (2) /home/mxnet/python/mxnet/../../lib/libmxnet.so(ZNSt17_Function_handlerIFN4nnvm5GraphES1_EPS2_E9_M_invokeERKSt9_Any_dataOS1+0x111) [0x7f28b2c033f1]
[bt] (3) /home/mxnet/python/mxnet/../../lib/libmxnet.so(_ZN4nnvm11ApplyPassesENS_5GraphERKSt6vectorINSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEESaIS7_EE+0x32c) [0x7f28b4053b8c]
[bt] (4) /home/mxnet/python/mxnet/../../lib/libmxnet.so(_ZN4nnvm9ApplyPassENS_5GraphERKNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEE+0x3c9) [0x7f28b2f99389]
[bt] (5) /home/mxnet/python/mxnet/../../lib/libmxnet.so(_ZN4nnvm4pass8GradientENS_5GraphESt6vectorINS_9NodeEntryESaIS3_EES5_S5_St8functionIFS3_OS5_EES6_IFiRKNS_4NodeEEES6_IFS3_RKS3_SG_EES2_IPKNS_2OpESaISL_EENSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEE+0x60c) [0x7f28b300780c]
[bt] (6) /home/mxnet/python/mxnet/../../lib/libmxnet.so(_ZN5mxnet4exec13GraphExecutor13InitFullGraphEN4nnvm6SymbolERKSt6vectorINS_9OpReqTypeESaIS5_EE+0x863) [0x7f28b2ff1283]
[bt] (7) /home/mxnet/python/mxnet/../../lib/libmxnet.so(_ZN5mxnet4exec13GraphExecutor9InitGraphEN4nnvm6SymbolERKNS_7ContextERKSt3mapINSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEES4_St4lessISD_ESaISt4pairIKSD_S4_EEERKSt6vectorIS4_SaIS4_EESR_SR_RKSN_INS_9OpReqTypeESaISS_EE+0x82) [0x7f28b2ff1b52]
[bt] (8) /home/mxnet/python/mxnet/../../lib/libmxnet.so(_ZN5mxnet4exec13GraphExecutor4InitEN4nnvm6SymbolERKNS_7ContextERKSt3mapINSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEES4_St4lessISD_ESaISt4pairIKSD_S4_EEERKSt6vectorINS_7NDArrayESaISO_EESS_RKSN_INS_9OpReqTypeESaIST_EESS_PNS_8ExecutorERKSt13unordered_mapINS2_9NodeEntryESO_NS2_13NodeEntryHashENS2_14NodeEntryEqualESaISG_IKS11_SO_EEE+0x76f) [0x7f28b2ffc45f]
[bt] (9) /home/mxnet/python/mxnet/../../lib/libmxnet.so(ZN5mxnet8Executor4BindEN4nnvm6SymbolERKNS_7ContextERKSt3mapINSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEES3_St4lessISC_ESaISt4pairIKSC_S3_EEERKSt6vectorINS_7NDArrayESaISN_EESR_RKSM_INS_9OpReqTypeESaISS_EESR_PS0+0x1b8) [0x7f28b2ffd7f8]
`

初始化center向量

初始化的每个类的center向量是多少？0向量么？

train error: mx.symbol.Custom type error

运行train.py 报错：

TypeError: init() got an unexpected keyword argument 'data'

报错代码行

center_loss_ = mx.symbol.Custom(
        data=net, label=center_label, name='center_loss_', op_type='centerloss',
        num_class=NUM_CLASS, alpha=0.5, scale=1.0, batchsize=batch_size)

报错信息

类别中心center参数的另一种方式

能否将类别中心作为一个网络变量，center loss operator能否输入接收一个centor变量，输出时更新该变量，这样不同卡之间就能保持一致了。不知这样是否具有可行性？谢谢

run vis.py 时出现错误

C:\WinPython\python-2.7.10.amd64\python.exe D:/MyCoding/DeepLearning/test_example/mxnet/example/center_loss/vis.py
feature_extractor loaded
[01:15:43] D:\MyCoding\DeepLearning\OpenSource\mxnet\src\io\iter_mnist.cc:94: MNISTIter: load 60000 images, shuffle=1, shape=(100,1,28,28)
[01:15:44] D:\MyCoding\DeepLearning\OpenSource\mxnet\src\io\iter_mnist.cc:94: MNISTIter: load 10000 images, shuffle=1, shape=(100,1,28,28)
extracting feature
[01:15:44] D:\MyCoding\DeepLearning\OpenSource\mxnet\dmlc-core\include\dmlc/logging.h:235: [01:15:44] D:\MyCoding\DeepLearning\OpenSource\mxnet\src\symbol\symbol.cc:155: Symbol.InferShapeKeyword argument name softmax_label not found.
Candidate arguments:
[0]data
[1]convolution0_weight
[2]convolution0_bias
[3]convolution1_weight
[4]convolution1_bias
[5]fullyconnected0_weight
[6]fullyconnected0_bias
[7]embedding_weight
[8]embedding_bias

Traceback (most recent call last):
File "D:/MyCoding/DeepLearning/test_example/mxnet/example/center_loss/vis.py", line 73, in
main()
File "D:/MyCoding/DeepLearning/test_example/mxnet/example/center_loss/vis.py", line 55, in main
preds = feature_extractor.predict( i.data[0] )
File "C:\WinPython\python-2.7.10.amd64\lib\site-packages\mxnet-0.7.0-py2.7.egg\mxnet\model.py", line 608, in predict
self._init_predictor(data_shapes, type_dict)
File "C:\WinPython\python-2.7.10.amd64\lib\site-packages\mxnet-0.7.0-py2.7.egg\mxnet\model.py", line 532, in _init_predictor
self.ctx[0], grad_req='null', type_dict=type_dict, **dict(input_shapes))
File "C:\WinPython\python-2.7.10.amd64\lib\site-packages\mxnet-0.7.0-py2.7.egg\mxnet\symbol.py", line 676, in simple_bind
arg_shapes, _, aux_shapes = self.infer_shape(**kwargs)
File "C:\WinPython\python-2.7.10.amd64\lib\site-packages\mxnet-0.7.0-py2.7.egg\mxnet\symbol.py", line 458, in infer_shape
return self._infer_shape_impl(False, *args, **kwargs)
File "C:\WinPython\python-2.7.10.amd64\lib\site-packages\mxnet-0.7.0-py2.7.egg\mxnet\symbol.py", line 518, in _infer_shape_impl
ctypes.byref(complete)))
File "C:\WinPython\python-2.7.10.amd64\lib\site-packages\mxnet-0.7.0-py2.7.egg\mxnet\base.py", line 77, in check_call
raise MXNetError(py_str(_LIB.MXGetLastError()))
mxnet.base.MXNetError: [01:15:44] D:\MyCoding\DeepLearning\OpenSource\mxnet\src\symbol\symbol.cc:155: Symbol.InferShapeKeyword argument name softmax_label not found.
Candidate arguments:
[0]data
[1]convolution0_weight
[2]convolution0_bias
[3]convolution1_weight
[4]convolution1_bias
[5]fullyconnected0_weight
[6]fullyconnected0_bias
[7]embedding_weight
[8]embedding_bias