tensorflow object detection api graph rewriter

目前,tensorflow 目標識別的api函數可使用 graph rewriter這樣的配置,這樣配置的引入主要是爲了模型壓縮使用,具體設置參數有:

syntax = "proto2";

package object_detection.protos;

// Message to configure graph rewriter for the tf graph.
message GraphRewriter {
  optional Quantization quantization = 1;
}

// Message for quantization options. See
// tensorflow/contrib/quantize/python/quantize.py for details.
message Quantization {
  // Number of steps to delay before quantization takes effect during training.
  optional int32 delay = 1 [default = 500000];

  // Number of bits to use for quantizing weights.
  // Only 8 bit is supported for now.
  optional int32 weight_bits = 2 [default = 8];

  // Number of bits to use for quantizing activations.
  // Only 8 bit is supported for now.
  optional int32 activation_bits = 3 [default = 8];
}


實際在pipline.config裏面是:

graph_rewriter {
  quantization {
    delay: 48000  # 迭代次數後使用graph_rewriter 量化

    activation_bits: 8 #激活位數

    weight_bits: 8 #權重位數,支持int8  }}