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.

aipp.h 2.4 kB

5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172
  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. /*!
  17. * \file aipp.h
  18. * \brief
  19. */
  20. #ifndef GE_OP_AIPP_H
  21. #define GE_OP_AIPP_H
  22. #include "graph/operator_reg.h"
  23. namespace ge {
  24. /**
  25. *@brief Performs AI pre-processing (AIPP) on images including color space conversion (CSC), image normalization (by subtracting the mean value or multiplying a factor), image cropping (by specifying the crop start and cropping the image to the size required by the neural network), and much more.
  26. *@par Inputs:
  27. *@li images: An NCHW or NHWC tensor of type uint8, specifying the input to the data layer.
  28. *@li params: Dynamic AIPP configuration parameters of type uint8.
  29. *@par Attributes:
  30. *aipp_config_path: A required string, specifying the path of the AIPP configuration file
  31. *@par Outputs:
  32. *features: The AIPP-processed output tensor of type float16 or uint8.
  33. *@par Third-party framework compatibility
  34. * It is a custom operator. It has no corresponding operator in Caffe.
  35. */
  36. REG_OP(Aipp)
  37. .INPUT(images, TensorType{DT_UINT8})
  38. .OPTIONAL_INPUT(params, TensorType{DT_UINT8})
  39. .OUTPUT(features, TensorType({DT_FLOAT16, DT_UINT8}))
  40. .ATTR(aipp_config_path, String, "./aipp.cfg")
  41. .OP_END_FACTORY_REG(Aipp)
  42. /**
  43. *@brief Performs this op is for dynamic aipp.If you set aipp-mode to dynamic \n
  44. in aipp config file, framework will auto add one input node to graph at last.
  45. *@par Inputs:
  46. *data: An NCHW or NHWC tensor of type uint8, specifying the input to the data layer.
  47. *@par Attributes:
  48. *index: specify aipp serial num
  49. *@par Outputs:
  50. *out: The AIPP-processed output tensor of all types.
  51. *@par Third-party framework compatibility
  52. *Compatible with the TensorFlow operator AippData.
  53. */
  54. REG_OP(AippData)
  55. .INPUT(data, TensorType::ALL())
  56. .OUTPUT(out, TensorType::ALL())
  57. .ATTR(index, Int, 0)
  58. .OP_END_FACTORY_REG(AippData)
  59. } // namespace ge
  60. #endif // GE_OP_AIPP_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