reid from https://github.com/michuanhaohao/reid-strong-baseline
zhangmeng
2020-01-17 f7c4a3cfd07adede3308f8d9d3d7315427d90a7c
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
// Copyright 2004-present Facebook. All Rights Reserved.
 
#pragma once
 
#include "caffe2/core/context.h"
#include "caffe2/core/logging.h"
#include "caffe2/core/operator.h"
#include "caffe2/utils/math.h"
 
namespace caffe2 {
 
template <typename T, class Context>
class LambdaRankNdcgOp final : public Operator<Context> {
 public:
  template <class... Args>
  explicit LambdaRankNdcgOp(Args&&... args)
      : Operator<Context>(std::forward<Args>(args)...),
        use_ndcg_as_loss_(
            this->template GetSingleArgument<bool>("use_ndcg_as_loss", false)),
        use_idcg_normalization_(this->template GetSingleArgument<bool>(
            "use_idcg_normalization",
            true)),
        use_exp_gain_(
            this->template GetSingleArgument<bool>("use_exp_gain", true)) {}
  USE_OPERATOR_CONTEXT_FUNCTIONS;
  bool RunOnDevice() override;
 
 private:
  INPUT_TAGS(PRED, REL, SESSION_LENS);
  OUTPUT_TAGS(LOSS, DPRED);
 
  void ResizeInvLogITensor(int);
  void ComputeDiscounts(int*, int);
  float LambdaRankNdcgSession(
      int start_index,
      int end_index,
      const Tensor& y,
      const Tensor& r,
      Tensor** dy);
  bool use_ndcg_as_loss_;
  bool use_idcg_normalization_;
  bool use_exp_gain_;
  Tensor gain_;
  Tensor discount_;
  Tensor rank_idx_;
  Tensor ideal_idx_;
  Tensor lambda_;
  Tensor inv_log_i_;
};
 
template <typename T, class Context>
class LambdaRankNdcgGradientOp final : public Operator<Context> {
 public:
  USE_SIMPLE_CTOR_DTOR(LambdaRankNdcgGradientOp);
  USE_OPERATOR_CONTEXT_FUNCTIONS;
  bool RunOnDevice() override;
 
 private:
  INPUT_TAGS(Y, SESSION_LENS, DY_CACHE, DLOSS);
  OUTPUT_TAGS(DY);
};
 
} // namespace caffe2