ctc_align_op.cc 3.0 KB
Newer Older
1
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.
W
wanghaoshuang 已提交
2 3 4 5 6 7 8 9 10 11 12 13 14

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. */

Y
Yi Wang 已提交
15
#include "paddle/fluid/operators/ctc_align_op.h"
W
wanghaoshuang 已提交
16 17 18 19

namespace paddle {
namespace operators {

W
wanghaoshuang 已提交
20
class CTCAlignOp : public framework::OperatorWithKernel {
W
wanghaoshuang 已提交
21 22 23 24 25
 public:
  using framework::OperatorWithKernel::OperatorWithKernel;

  void InferShape(framework::InferShapeContext* ctx) const override {
    PADDLE_ENFORCE(ctx->HasInput("Input"),
W
wanghaoshuang 已提交
26
                   "Input of CTCAlignOp should not be null.");
W
wanghaoshuang 已提交
27
    PADDLE_ENFORCE(ctx->HasOutput("Output"),
W
wanghaoshuang 已提交
28
                   "Output of CTCAlignOp should not be null.");
W
wanghaoshuang 已提交
29 30 31 32

    auto input_dims = ctx->GetInputDim("Input");

    // TODO(wanghaoshuang): it is tricky to set the wrong dimension here.
33
    ctx->SetOutputDim("Output", input_dims);
W
wanghaoshuang 已提交
34 35 36 37 38 39 40 41 42 43 44
  }

 protected:
  framework::OpKernelType GetExpectedKernelType(
      const framework::ExecutionContext& ctx) const override {
    return framework::OpKernelType(
        framework::ToDataType(ctx.Input<Tensor>("Input")->type()),
        ctx.device_context());
  }
};

W
wanghaoshuang 已提交
45
class CTCAlignOpMaker : public framework::OpProtoAndCheckerMaker {
W
wanghaoshuang 已提交
46
 public:
Y
Yu Yang 已提交
47
  void Make() override {
W
wanghaoshuang 已提交
48
    AddInput("Input",
49 50
             "(LodTensor, default: LoDTensor<int>), Its shape is "
             "[Lp, 1], where Lp is the sum of all input sequences' length.");
W
wanghaoshuang 已提交
51
    AddOutput("Output", "(Tensor, default: Tensor<int>), The align result.");
W
wanghaoshuang 已提交
52 53
    AddAttr<int>("blank",
                 "(int, default: 0), the blank label setted in Connectionist "
54
                 "Temporal Classification (CTC) op.")
W
wanghaoshuang 已提交
55 56 57 58 59 60
        .SetDefault(0);
    AddAttr<bool>("merge_repeated",
                  "(bool, default: true), whether to "
                  "merge repeated elements between two blanks. ")
        .SetDefault(true);
    AddComment(R"DOC(
W
wanghaoshuang 已提交
61
CTCAlign op is used to merge repeated elements between two blanks
62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79
and then delete all blanks in sequence.

Given:
    Input.data = [0, 1, 2, 2, 0, 4, 0, 4, 5, 0, 6,
                  6, 0, 0, 7, 7, 7, 0]
    Input.dims = {18, 1}
    Input.LoD = [[0, 11, 18]]

And:
    blank = 0
    merge_repeated = True

Then:
    Output.data = [1, 2, 4, 4, 5, 6,
                   6, 7]
    Output.dims = {8, 1}
    Output.LoD = [[0, 6, 8]]

W
wanghaoshuang 已提交
80 81 82 83 84 85 86 87
)DOC");
  }
};

}  // namespace operators
}  // namespace paddle

namespace ops = paddle::operators;
W
wanghaoshuang 已提交
88
REGISTER_OPERATOR(ctc_align, ops::CTCAlignOp, ops::CTCAlignOpMaker,
W
wanghaoshuang 已提交
89 90
                  paddle::framework::EmptyGradOpMaker);
REGISTER_OP_CPU_KERNEL(
W
wanghaoshuang 已提交
91 92
    ctc_align, ops::CTCAlignKernel<paddle::platform::CPUDeviceContext, int>,
    ops::CTCAlignKernel<paddle::platform::CPUDeviceContext, int64_t>);