Created by: XipengY

• faster rcnn 架构 修改的shuffle net backbone
• 平台：centros6.3
• paddle版本号：1.5.0
• GPU：K40
• python2.7

训练信息

1）单机/多机

reshape的时候，维度对不上： 注：训练了大概10多个iter报错的，不是一开始就报错了。

错误信息如下：

Python Callstacks:
  File "/lib/python2.7/site-packages/paddle/fluid/framework.py", line 1748, in append_op
    attrs=kwargs.get("attrs", None))
  File "lib/python2.7/site-packages/paddle/fluid/layer_helper.py", line 43, in append_op
    return self.main_program.current_block().append_op(*args, **kwargs)
  File "lib/python2.7/site-packages/paddle/fluid/layers/nn.py", line 6797, in reshape
    "XShape": x_shape})
  File "Paddle_models/PaddleCV/PaddleDetection/ppdet/modeling/backbones/snet.py", line 270, in channel_shuffle
    x = fluid.layers.reshape(x=x, shape=[batchsize, groups, channels_per_group, height, width])
  File "PaddleCV/PaddleDetection/ppdet/modeling/backbones/snet.py", line 380, in inverted_residual_unit
    return self.channel_shuffle(out, 2)
  File "PaddleCV/PaddleDetection/ppdet/modeling/backbones/snet.py", line 106, in __call__
    benchmodel=2,name=str(idxstage+2)+'_'+str(i+1))
  File "PaddleCV/PaddleDetection/ppdet/modeling/architectures/faster_rcnn_thunder.py", line 70, in build
    body_feats = self.backbone(im)
  File "PaddleCV/PaddleDetection/ppdet/modeling/architectures/faster_rcnn_thunder.py", line 125, in train
    return self.build(feed_vars, 'train')
  File "tools/train.py", line 102, in main
    train_fetches = model.train(feed_vars)
  File "tools/train.py", line 234, in <module>
    main()
C++ Callstacks:
Enforce failed. Expected output_shape[unk_dim_idx] * capacity == -in_size, but received output_shape[unk_dim_idx] * capacity:-384000 != -in_size:-432000.
Invalid shape is given. at [/paddle/paddle/fluid/operators/reshape_op.cc:131]

组网中打印的维度和预想的一致：

name: "concat_0.tmp_0"
type {
  type: LOD_TENSOR
  lod_tensor {
    tensor {
      data_type: FP32
      dims: -1
      dims: 60
      dims: 40
      dims: 40
    }
    lod_level: 0
  }
}

但是训练中打印了tensor，确实有奇怪的现象：

Tensor[concat_0.tmp_0]
        shape: [6,60,33,40,]
        dtype: f