multiclass_nms_op.cc 15.2 KB
Newer Older
D
dangqingqing 已提交
1
/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve.
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/framework/op_registry.h"
16 17 18 19 20 21 22 23 24 25

namespace paddle {
namespace operators {

using Tensor = framework::Tensor;
using LoDTensor = framework::LoDTensor;

constexpr int64_t kOutputDim = 6;
constexpr int64_t kBBoxSize = 4;

D
dangqingqing 已提交
26
class MultiClassNMSOp : public framework::OperatorWithKernel {
27 28 29 30
 public:
  using framework::OperatorWithKernel::OperatorWithKernel;

  void InferShape(framework::InferShapeContext* ctx) const override {
D
dangqingqing 已提交
31 32
    PADDLE_ENFORCE(ctx->HasInput("BBoxes"),
                   "Input(BBoxes) of MultiClassNMS should not be null.");
33
    PADDLE_ENFORCE(ctx->HasInput("Scores"),
D
dangqingqing 已提交
34 35 36
                   "Input(Scores) of MultiClassNMS should not be null.");
    PADDLE_ENFORCE(ctx->HasOutput("Out"),
                   "Output(Out) of MultiClassNMS should not be null.");
37

D
dangqingqing 已提交
38
    auto box_dims = ctx->GetInputDim("BBoxes");
39 40
    auto score_dims = ctx->GetInputDim("Scores");

41 42
    PADDLE_ENFORCE_EQ(box_dims.size(), 3,
                      "The rank of Input(BBoxes) must be 3.");
43 44
    PADDLE_ENFORCE_EQ(score_dims.size(), 3,
                      "The rank of Input(Scores) must be 3.");
45
    PADDLE_ENFORCE_EQ(box_dims[2], 4,
D
dangqingqing 已提交
46 47 48
                      "The 2nd dimension of Input(BBoxes) must be 4, "
                      "represents the layout of coordinate "
                      "[xmin, ymin, xmax, ymax]");
49
    PADDLE_ENFORCE_EQ(box_dims[1], score_dims[2],
D
dangqingqing 已提交
50 51 52
                      "The 1st dimensiong of Input(BBoxes) must be equal to "
                      "3rd dimension of Input(Scores), which represents the "
                      "predicted bboxes.");
53 54 55

    // Here the box_dims[0] is not the real dimension of output.
    // It will be rewritten in the computing kernel.
56
    ctx->SetOutputDim("Out", {box_dims[1], 6});
57
  }
D
dangqingqing 已提交
58 59 60 61 62 63 64 65 66

 protected:
  framework::OpKernelType GetExpectedKernelType(
      const framework::ExecutionContext& ctx) const override {
    return framework::OpKernelType(
        framework::ToDataType(
            ctx.Input<framework::LoDTensor>("Scores")->type()),
        ctx.device_context());
  }
67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87
};

template <class T>
bool SortScorePairDescend(const std::pair<float, T>& pair1,
                          const std::pair<float, T>& pair2) {
  return pair1.first > pair2.first;
}

template <class T>
static inline void GetMaxScoreIndex(
    const std::vector<T>& scores, const T threshold, int top_k,
    std::vector<std::pair<T, int>>* sorted_indices) {
  for (size_t i = 0; i < scores.size(); ++i) {
    if (scores[i] > threshold) {
      sorted_indices->push_back(std::make_pair(scores[i], i));
    }
  }
  // Sort the score pair according to the scores in descending order
  std::stable_sort(sorted_indices->begin(), sorted_indices->end(),
                   SortScorePairDescend<int>);
  // Keep top_k scores if needed.
88
  if (top_k > -1 && top_k < static_cast<int>(sorted_indices->size())) {
89 90 91 92 93
    sorted_indices->resize(top_k);
  }
}

template <class T>
94
static inline T BBoxArea(const T* box, const bool normalized) {
95
  if (box[2] < box[0] || box[3] < box[1]) {
D
dangqingqing 已提交
96 97 98
    // If coordinate values are is invalid
    // (e.g. xmax < xmin or ymax < ymin), return 0.
    return static_cast<T>(0.);
99 100 101 102 103 104
  } else {
    const T w = box[2] - box[0];
    const T h = box[3] - box[1];
    if (normalized) {
      return w * h;
    } else {
D
dangqingqing 已提交
105
      // If coordinate values are not within range [0, 1].
106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131
      return (w + 1) * (h + 1);
    }
  }
}

template <class T>
static inline T JaccardOverlap(const T* box1, const T* box2,
                               const bool normalized) {
  if (box2[0] > box1[2] || box2[2] < box1[0] || box2[1] > box1[3] ||
      box2[3] < box1[1]) {
    return static_cast<T>(0.);
  } else {
    const T inter_xmin = std::max(box1[0], box2[0]);
    const T inter_ymin = std::max(box1[1], box2[1]);
    const T inter_xmax = std::min(box1[2], box2[2]);
    const T inter_ymax = std::min(box1[3], box2[3]);
    const T inter_w = inter_xmax - inter_xmin;
    const T inter_h = inter_ymax - inter_ymin;
    const T inter_area = inter_w * inter_h;
    const T bbox1_area = BBoxArea<T>(box1, normalized);
    const T bbox2_area = BBoxArea<T>(box2, normalized);
    return inter_area / (bbox1_area + bbox2_area - inter_area);
  }
}

template <typename T>
D
dangqingqing 已提交
132
class MultiClassNMSKernel : public framework::OpKernel<T> {
133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153
 public:
  void NMSFast(const Tensor& bbox, const Tensor& scores,
               const T score_threshold, const T nms_threshold, const T eta,
               const int64_t top_k, std::vector<int>* selected_indices) const {
    // The total boxes for each instance.
    int64_t num_boxes = bbox.dims()[0];
    // 4: [xmin ymin xmax ymax]
    int64_t box_size = bbox.dims()[1];

    std::vector<T> scores_data(num_boxes);
    std::copy_n(scores.data<T>(), num_boxes, scores_data.begin());
    std::vector<std::pair<T, int>> sorted_indices;
    GetMaxScoreIndex(scores_data, score_threshold, top_k, &sorted_indices);

    selected_indices->clear();
    T adaptive_threshold = nms_threshold;
    const T* bbox_data = bbox.data<T>();

    while (sorted_indices.size() != 0) {
      const int idx = sorted_indices.front().second;
      bool keep = true;
154
      for (size_t k = 0; k < selected_indices->size(); ++k) {
155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173
        if (keep) {
          const int kept_idx = (*selected_indices)[k];
          T overlap = JaccardOverlap<T>(bbox_data + idx * box_size,
                                        bbox_data + kept_idx * box_size, true);
          keep = overlap <= adaptive_threshold;
        } else {
          break;
        }
      }
      if (keep) {
        selected_indices->push_back(idx);
      }
      sorted_indices.erase(sorted_indices.begin());
      if (keep && eta < 1 && adaptive_threshold > 0.5) {
        adaptive_threshold *= eta;
      }
    }
  }

D
dangqingqing 已提交
174
  void MultiClassNMS(const framework::ExecutionContext& ctx,
175
                     const Tensor& scores, const Tensor& bboxes,
D
dangqingqing 已提交
176 177
                     std::map<int, std::vector<int>>& indices,
                     int& num_nmsed_out) const {
D
dangqingqing 已提交
178 179 180
    int64_t background_label = ctx.Attr<int>("background_label");
    int64_t nms_top_k = ctx.Attr<int>("nms_top_k");
    int64_t keep_top_k = ctx.Attr<int>("keep_top_k");
181 182
    T nms_threshold = static_cast<T>(ctx.Attr<float>("nms_threshold"));
    T nms_eta = static_cast<T>(ctx.Attr<float>("nms_eta"));
D
dangqingqing 已提交
183
    T score_threshold = static_cast<T>(ctx.Attr<float>("score_threshold"));
184 185 186 187 188 189 190 191

    int64_t class_num = scores.dims()[0];
    int64_t predict_dim = scores.dims()[1];
    int num_det = 0;
    for (int64_t c = 0; c < class_num; ++c) {
      if (c == background_label) continue;
      Tensor score = scores.Slice(c, c + 1);
      NMSFast(bboxes, score, score_threshold, nms_threshold, nms_eta, nms_top_k,
D
dangqingqing 已提交
192 193
              &(indices[c]));
      num_det += indices[c].size();
194 195
    }

D
dangqingqing 已提交
196
    num_nmsed_out = num_det;
197 198 199
    const T* scores_data = scores.data<T>();
    if (keep_top_k > -1 && num_det > keep_top_k) {
      std::vector<std::pair<float, std::pair<int, int>>> score_index_pairs;
D
dangqingqing 已提交
200
      for (const auto& it : indices) {
201 202 203
        int label = it.first;
        const T* sdata = scores_data + label * predict_dim;
        const std::vector<int>& label_indices = it.second;
204
        for (size_t j = 0; j < label_indices.size(); ++j) {
205 206 207 208 209 210 211
          int idx = label_indices[j];
          PADDLE_ENFORCE_LT(idx, predict_dim);
          score_index_pairs.push_back(
              std::make_pair(sdata[idx], std::make_pair(label, idx)));
        }
      }
      // Keep top k results per image.
212 213
      std::stable_sort(score_index_pairs.begin(), score_index_pairs.end(),
                       SortScorePairDescend<std::pair<int, int>>);
214 215 216 217
      score_index_pairs.resize(keep_top_k);

      // Store the new indices.
      std::map<int, std::vector<int>> new_indices;
218
      for (size_t j = 0; j < score_index_pairs.size(); ++j) {
219 220 221 222
        int label = score_index_pairs[j].second.first;
        int idx = score_index_pairs[j].second.second;
        new_indices[label].push_back(idx);
      }
D
dangqingqing 已提交
223 224
      new_indices.swap(indices);
      num_nmsed_out = keep_top_k;
225 226 227
    }
  }

D
dangqingqing 已提交
228
  void MultiClassOutput(const Tensor& scores, const Tensor& bboxes,
229 230 231 232 233 234 235 236 237 238 239
                        std::map<int, std::vector<int>>& selected_indices,
                        Tensor* outs) const {
    int predict_dim = scores.dims()[1];
    auto* scores_data = scores.data<T>();
    auto* bboxes_data = bboxes.data<T>();
    auto* odata = outs->data<T>();

    int count = 0;
    for (const auto& it : selected_indices) {
      int label = it.first;
      const T* sdata = scores_data + label * predict_dim;
D
dangqingqing 已提交
240
      const std::vector<int>& indices = it.second;
241
      for (size_t j = 0; j < indices.size(); ++j) {
242 243 244 245
        int idx = indices[j];
        const T* bdata = bboxes_data + idx * kBBoxSize;
        odata[count * kOutputDim] = label;           // label
        odata[count * kOutputDim + 1] = sdata[idx];  // score
D
dangqingqing 已提交
246 247
        // xmin, ymin, xmax, ymax
        std::memcpy(odata + count * kOutputDim + 2, bdata, 4 * sizeof(T));
D
dangqingqing 已提交
248
        count++;
249 250 251 252 253
      }
    }
  }

  void Compute(const framework::ExecutionContext& ctx) const override {
D
dangqingqing 已提交
254
    auto* boxes = ctx.Input<Tensor>("BBoxes");
255 256 257 258 259
    auto* scores = ctx.Input<Tensor>("Scores");
    auto* outs = ctx.Output<LoDTensor>("Out");

    auto score_dims = scores->dims();

D
dangqingqing 已提交
260
    int64_t batch_size = score_dims[0];
261 262
    int64_t class_num = score_dims[1];
    int64_t predict_dim = score_dims[2];
263
    int64_t box_dim = boxes->dims()[2];
264 265 266 267 268 269

    std::vector<std::map<int, std::vector<int>>> all_indices;
    std::vector<size_t> batch_starts = {0};
    for (int64_t i = 0; i < batch_size; ++i) {
      Tensor ins_score = scores->Slice(i, i + 1);
      ins_score.Resize({class_num, predict_dim});
270 271 272 273

      Tensor ins_boxes = boxes->Slice(i, i + 1);
      ins_boxes.Resize({predict_dim, box_dim});

274 275
      std::map<int, std::vector<int>> indices;
      int num_nmsed_out = 0;
276
      MultiClassNMS(ctx, ins_score, ins_boxes, indices, num_nmsed_out);
277 278 279 280 281 282
      all_indices.push_back(indices);
      batch_starts.push_back(batch_starts.back() + num_nmsed_out);
    }

    int num_kept = batch_starts.back();
    if (num_kept == 0) {
283 284
      T* od = outs->mutable_data<T>({1}, ctx.GetPlace());
      od[0] = -1;
285 286 287 288 289
    } else {
      outs->mutable_data<T>({num_kept, kOutputDim}, ctx.GetPlace());
      for (int64_t i = 0; i < batch_size; ++i) {
        Tensor ins_score = scores->Slice(i, i + 1);
        ins_score.Resize({class_num, predict_dim});
290 291 292 293

        Tensor ins_boxes = boxes->Slice(i, i + 1);
        ins_boxes.Resize({predict_dim, box_dim});

294 295 296 297
        int64_t s = batch_starts[i];
        int64_t e = batch_starts[i + 1];
        if (e > s) {
          Tensor out = outs->Slice(s, e);
298
          MultiClassOutput(ins_score, ins_boxes, all_indices[i], &out);
299 300 301 302 303 304 305 306 307 308 309
        }
      }
    }

    framework::LoD lod;
    lod.emplace_back(batch_starts);

    outs->set_lod(lod);
  }
};

D
dangqingqing 已提交
310
class MultiClassNMSOpMaker : public framework::OpProtoAndCheckerMaker {
311
 public:
D
dangqingqing 已提交
312
  MultiClassNMSOpMaker(OpProto* proto, OpAttrChecker* op_checker)
313
      : OpProtoAndCheckerMaker(proto, op_checker) {
D
dangqingqing 已提交
314
    AddInput("BBoxes",
315 316 317
             "(Tensor) A 3-D Tensor with shape [N, M, 4] represents the "
             "predicted locations of M bounding bboxes, N is the batch size. "
             "Each bounding box has four coordinate values and the layout is "
D
dangqingqing 已提交
318
             "[xmin, ymin, xmax, ymax].");
D
dangqingqing 已提交
319 320
    AddInput("Scores",
             "(Tensor) A 3-D Tensor with shape [N, C, M] represents the "
D
dangqingqing 已提交
321 322 323 324
             "predicted confidence predictions. N is the batch size, C is the "
             "class number, M is number of bounding boxes. For each category "
             "there are total M scores which corresponding M bounding boxes. "
             " Please note, M is equal to the 1st dimension of BBoxes. ");
D
dangqingqing 已提交
325
    AddAttr<int>(
326 327
        "background_label",
        "(int64_t, defalut: 0) "
D
dangqingqing 已提交
328 329
        "The index of background label, the background label will be ignored. "
        "If set to -1, then all categories will be considered.")
330
        .SetDefault(0);
D
dangqingqing 已提交
331 332
    AddAttr<float>("score_threshold",
                   "(float) "
D
dangqingqing 已提交
333 334
                   "Threshold to filter out bounding boxes with low "
                   "confidence score. If not provided, consider all boxes.");
D
dangqingqing 已提交
335 336 337 338 339
    AddAttr<int>("nms_top_k",
                 "(int64_t) "
                 "Maximum number of detections to be kept according to the "
                 "confidences aftern the filtering detections based on "
                 "score_threshold");
340 341
    AddAttr<float>("nms_threshold",
                   "(float, defalut: 0.3) "
D
dangqingqing 已提交
342
                   "The threshold to be used in NMS.")
343 344 345
        .SetDefault(0.3);
    AddAttr<float>("nms_eta",
                   "(float) "
D
dangqingqing 已提交
346
                   "The parameter for adaptive NMS.")
347
        .SetDefault(1.0);
D
dangqingqing 已提交
348 349 350 351
    AddAttr<int>("keep_top_k",
                 "(int64_t) "
                 "Number of total bboxes to be kept per image after NMS "
                 "step. -1 means keeping all bboxes after NMS step.");
352 353 354 355 356 357 358 359 360
    AddOutput("Out",
              "(LoDTensor) A 2-D LoDTensor with shape [No, 6] represents the "
              "detections. Each row has 6 values: "
              "[label, confidence, xmin, ymin, xmax, ymax], No is the total "
              "number of detections in this mini-batch. For each instance, "
              "the offsets in first dimension are called LoD, the number of "
              "offset is N + 1, if LoD[i + 1] - LoD[i] == 0, means there is "
              "no detected bbox.");
    AddComment(R"DOC(
D
dangqingqing 已提交
361
This operator is to do multi-class non maximum suppression (NMS) on a batched
362 363
of boxes and scores.

D
dangqingqing 已提交
364 365 366 367 368 369 370
In the NMS step, this operator greedily selects a subset of detection bounding
boxes that have high scores larger than score_threshold, if providing this
threshold, then selects the largest nms_top_k confidences scores if nms_top_k
is larger than -1. Then this operator pruns away boxes that have high IOU
(intersection over union) overlap with already selected boxes by adaptive
threshold NMS based on parameters of nms_threshold and nms_eta.

371
Aftern NMS step, at most keep_top_k number of total bboxes are to be kept
D
dangqingqing 已提交
372
per image if keep_top_k is larger than -1.
373

D
dangqingqing 已提交
374
This operator support multi-class and batched inputs. It applying NMS
375 376 377 378 379 380
independently for each class. The outputs is a 2-D LoDTenosr, for each
image, the offsets in first dimension of LoDTensor are called LoD, the number
of offset is N + 1, where N is the batch size. If LoD[i + 1] - LoD[i] == 0,
means there is no detected bbox for this image. If there is no detected boxes
for all images, all the elements in LoD are 0, and the Out only contains one
value which is -1.
381 382 383 384 385 386 387 388
)DOC");
  }
};

}  // namespace operators
}  // namespace paddle

namespace ops = paddle::operators;
D
dangqingqing 已提交
389 390
REGISTER_OPERATOR(multiclass_nms, ops::MultiClassNMSOp,
                  ops::MultiClassNMSOpMaker,
391
                  paddle::framework::EmptyGradOpMaker);
D
dangqingqing 已提交
392 393
REGISTER_OP_CPU_KERNEL(multiclass_nms, ops::MultiClassNMSKernel<float>,
                       ops::MultiClassNMSKernel<double>);