You can not select more than 25 topics Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.

rpn_ops.h 2.0 kB

5 years ago
5 years ago
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657
  1. /**
  2. * Copyright 2019-2020 Huawei Technologies Co., Ltd
  3. *
  4. * Licensed under the Apache License, Version 2.0 (the "License");
  5. * you may not use this file except in compliance with the License.
  6. * You may obtain a copy of the License at
  7. *
  8. * http://www.apache.org/licenses/LICENSE-2.0
  9. *
  10. * Unless required by applicable law or agreed to in writing, software
  11. * distributed under the License is distributed on an "AS IS" BASIS,
  12. * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13. * See the License for the specific language governing permissions and
  14. * limitations under the License.
  15. */
  16. #ifndef GE_OP_RPN_OPS_H
  17. #define GE_OP_RPN_OPS_H
  18. #include "../graph/operator_reg.h"
  19. namespace ge {
  20. /**
  21. *@brief Iteratively removes lower scoring boxes which have an IoU greater than
  22. * iou_threshold with higher scoring box according to their
  23. * intersection-over-union (IoU).
  24. *@par Input:
  25. * @li box_scores: 2-D tensor with shape of [N, 8], including proposal boxes and
  26. * corresponding confidence scores.
  27. * @par Attributes:
  28. * @li iou_threshold: An optional float. The threshold for deciding whether boxes
  29. * overlap too much with respect to IOU.
  30. * @par Outputs:
  31. * @li selected_boxes: 2-D tensor with shape of [N,5], representing filtered
  32. * boxes including proposal boxes and corresponding confidence scores.
  33. * @li selected_idx: 1-D tensor with shape of [N], representing the index of
  34. * input proposal boxes.
  35. * @li selected_mask: 1-D tensor with shape of [N], the symbol judging whether
  36. * the output proposal boxes is valid.
  37. * @attention Constraints:
  38. * The 2nd-dim of input box_scores must be equal to 8.\n
  39. * Only supports 2864 input boxes at one time.\n
  40. */
  41. REG_OP(NMSWithMask)
  42. .INPUT(box_scores, TensorType({DT_FLOAT, DT_FLOAT16}))
  43. .OUTPUT(selected_boxes, TensorType({DT_FLOAT, DT_FLOAT16}))
  44. .OUTPUT(selected_idx, TensorType({DT_INT32}))
  45. .OUTPUT(selected_mask, TensorType({DT_UINT8}))
  46. .ATTR(iou_threshold, Float, 0.5)
  47. .OP_END_FACTORY_REG(NMSWithMask)
  48. } // namespace ge
  49. #endif // GE_OP_TRAINING_OPS_H

图引擎模块(GE)是MindSpore的一个子模块,其代码由C++实现,位于前端模块ME和底层硬件之间,起到承接作用。图引擎模块以ME下发的图作为输入,然后进行一系列的深度图优化操作,最后输出一张可以在底层硬件上高效运行的图。GE针对昇腾AI处理器的硬件结构特点,做了特定的优化工作,以此来充分发挥出昇腾AI处理器的强大算力。在进行模型训练/推理时,GE会被自动调用而用户并不感知。GE主要由GE API和GE Core两部分组成,详细的架构图如下所示