Softmax_label magically appears

[samskalicky]

trafficstars

Description

The Softmax/SoftmaxOutput operator has an optional label input that is enabled by setting the ignore_label or use_length: https://github.com/apache/incubator-mxnet/blob/master/src/operator/nn/softmax-inl.h#L616-L618 https://github.com/apache/incubator-mxnet/blob/v1.x/src/operator/softmax_output-inl.h#L61-L63

Many common image classification/CNN models use softmax, but dont use the softmax_label input meaningfully. Yet there is an actual input to the model for this.

Notice in ResNet models:

    {
      "op": "null",
      "param": {},
      "name": "softmax_label",
      "inputs": [],
      "backward_source_id": -1
    },
    {
      "op": "SoftmaxOutput",
      "param": {
        "grad_scale": "1",
        "ignore_label": "-1",
        "multi_output": "False",
        "normalization": "null",
        "out_grad": "False",
        "preserve_shape": "False",
        "use_ignore": "False"
      },
      "name": "softmax",
      "inputs": [[131, 0], [132, 0]],
      "backward_source_id": -1
    }

Error Message

Symbol/Module flow:

Traceback (most recent call last):
  File "test.py", line 13, in <module>
    mod.set_params(args, aux)
  File "/home/ubuntu/incubator-mxnet/python/mxnet/module/module.py", line 350, in set_params
    allow_extra=allow_extra)
  File "/home/ubuntu/incubator-mxnet/python/mxnet/module/module.py", line 309, in init_params
    _impl(desc, arr, arg_params)
  File "/home/ubuntu/incubator-mxnet/python/mxnet/module/module.py", line 300, in _impl
    raise RuntimeError("%s is not presented" % name)
RuntimeError: softmax_label is not presented

Gluon flow:

Traceback (most recent call last):
  File "test.py", line 5, in <module>
    'resnet-18-0000.params')
  File "/home/ubuntu/incubator-mxnet/python/mxnet/gluon/block.py", line 1409, in imports
    dtype_source='saved')
  File "/home/ubuntu/incubator-mxnet/python/mxnet/gluon/parameter.py", line 1020, in load
    ignore_extra, restore_prefix, filename, cast_dtype, dtype_source)
  File "/home/ubuntu/incubator-mxnet/python/mxnet/gluon/parameter.py", line 1057, in load_dict
    name[lprefix:], error_str, _brief_print_list(arg_dict.keys()))
AssertionError: Parameter 'softmax_label' is missing in file: resnet-18-0000.params, which contains parameters: 'stage1_unit1_sc_weight', 'stage2_unit1_bn2_gamma', 'bn_data_moving_var', ..., 'stage2_unit2_conv1_weight', 'stage1_unit2_bn1_gamma', 'stage2_unit2_bn1_moving_var', 'stage2_unit2_bn2_moving_var'. Please make sure source and target networks have the same prefix.For more info on naming, please see https://mxnet.io/api/python/docs/tutorials/packages/gluon/blocks/naming.html

To Reproduce

comment out the allow_missing=True to reproduce the error messages.

import mxnet as mx

path='http://data.mxnet.io/models/imagenet/'
[mx.test_utils.download(path+'resnet/18-layers/resnet-18-0000.params'),
 mx.test_utils.download(path+'resnet/18-layers/resnet-18-symbol.json')]
 
block = mx.gluon.nn.SymbolBlock.imports('resnet-18-symbol.json',
                                        ['data'],
                                        'resnet-18-0000.params',
                                        allow_missing=True)

sym, args, aux = mx.model.load_checkpoint('resnet-18',0)
mod = mx.mod.Module(symbol=sym, context=mx.cpu(), label_names=None)
mod.bind(for_training=False, data_shapes=[('data', (1,3,224,224))],
         label_shapes=mod._label_shapes)
mod.set_params(args, aux,
               allow_missing=True)

[xmfbit]

I have met the same problem, and how to fix it?

[alaaobeid]

any fix?

[samskalicky]

You can set allow_missing=True like in the examples above to avoid the issue.