caffe 源码学习笔记(3) Net

Jan 12, 2020 · 5 min read · caffe

Net 基本介绍

网络通过组成和自微分共同定义一个函数及其梯度。

网络是一些Layer组成的DAG,也就是有向无环图，在caffe中通常由prototxt定义．

比如

 1name: "LogReg"
 2layer {
 3  name: "mnist"
 4  type: "Data"
 5  top: "data"
 6  top: "label"
 7  data_param {
 8    source: "input_leveldb"
 9    batch_size: 64
10  }
11}
12layer {
13  name: "ip"
14  type: "InnerProduct"
15  bottom: "data"
16  top: "ip"
17  inner_product_param {
18    num_output: 2
19  }
20}
21layer {
22  name: "loss"
23  type: "SoftmaxWithLoss"
24  bottom: "ip"
25  bottom: "label"
26  top: "loss"
27}

定义了

网络结构

值得强调的是，caffe中网络结构的定义与实现是无关的,这点比pytorch之类的深度学习框架不知高到哪里去了，也是caffe作为部署端的事实标准的重要原因．

Net 实现细节

我们先看下NetParameter的proto

 1
 2    message NetParameter {
 3      optional string name = 1; // consider giving the network a name
 4      // DEPRECATED. See InputParameter. The input blobs to the network.
 5      repeated string input = 3;
 6      // DEPRECATED. See InputParameter. The shape of the input blobs.
 7      repeated BlobShape input_shape = 8;
 8
 9      // 4D input dimensions -- deprecated.  Use "input_shape" instead.
10      // If specified, for each input blob there should be four
11      // values specifying the num, channels, height and width of the input blob.
12      // Thus, there should be a total of (4 * #input) numbers.
13      repeated int32 input_dim = 4;
14
15      // Whether the network will force every layer to carry out backward operation.
16      // If set False, then whether to carry out backward is determined
17      // automatically according to the net structure and learning rates.
18      optional bool force_backward = 5 [default = false];
19      // The current "state" of the network, including the phase, level, and stage.
20      // Some layers may be included/excluded depending on this state and the states
21      // specified in the layers' include and exclude fields.
22      optional NetState state = 6;
23
24      // Print debugging information about results while running Net::Forward,
25      // Net::Backward, and Net::Update.
26      optional bool debug_info = 7 [default = false];
27
28      // The layers that make up the net.  Each of their configurations, including
29      // connectivity and behavior, is specified as a LayerParameter.
30      repeated LayerParameter layer = 100;  // ID 100 so layers are printed last.
31
32      // DEPRECATED: use 'layer' instead.
33      repeated V1LayerParameter layers = 2;
34    }
35

都很好理解，除了NetState不知道是个什么东西．．．

继续往下看，看到和NetState有关的两个proto定义如下:

 1
 2message NetState {
 3  optional Phase phase = 1 [default = TEST];
 4  optional int32 level = 2 [default = 0];
 5  repeated string stage = 3;
 6}
 7
 8message NetStateRule {
 9  // Set phase to require the NetState have a particular phase (TRAIN or TEST)
10  // to meet this rule.
11  optional Phase phase = 1;
12
13  // Set the minimum and/or maximum levels in which the layer should be used.
14  // Leave undefined to meet the rule regardless of level.
15  optional int32 min_level = 2;
16  optional int32 max_level = 3;
17
18  // Customizable sets of stages to include or exclude.
19  // The net must have ALL of the specified stages and NONE of the specified
20  // "not_stage"s to meet the rule.
21  // (Use multiple NetStateRules to specify conjunctions of stages.)
22  repeated string stage = 4;
23  repeated string not_stage = 5;
24}

参考　 [Discussion] How to specify all-in-one networks #3864 得知，这些参数的主要目的是为了解决部署和训练时，需要多个prototxt的问题．　

实际上，部署和训练的protoxt可能大部分都是一样的，只不过是输入，输出以及少部分layer有些差别．如果分为多个文件，会带来的问题就是，如果要修改一些共同的layer,就需要修改两次．

于是可以通过这些参数来控制网络结构，使得部署和训练使用相同的prototxt ，得到一个"all-in-one network"

接下来值得看的是网络的构建部分．

对于一个Net，会顺序构建每个Layer

对于每个Layer,大致分为AppendTop,AppendBottom，AppendParam　三部分．我们重点关注前两部分．

 1
 2// Helper for Net::Init: add a new bottom blob to the net.
 3template <typename Dtype>
 4int Net<Dtype>::AppendBottom(const NetParameter& param, const int layer_id,
 5    const int bottom_id, set<string>* available_blobs,
 6    map<string, int>* blob_name_to_idx) {
 7  const LayerParameter& layer_param = param.layer(layer_id);
 8  const string& blob_name = layer_param.bottom(bottom_id);
 9  if (available_blobs->find(blob_name) == available_blobs->end()) {
10    LOG(FATAL) << "Unknown bottom blob '" << blob_name << "' (layer '"
11               << layer_param.name() << "', bottom index " << bottom_id << ")";
12  }
13  const int blob_id = (*blob_name_to_idx)[blob_name];
14  LOG_IF(INFO, Caffe::root_solver())
15      << layer_names_[layer_id] << " <- " << blob_name;
16  bottom_vecs_[layer_id].push_back(blobs_[blob_id].get());
17  bottom_id_vecs_[layer_id].push_back(blob_id);
18  // 　？？？　怎么就erase了，一个top后面可以接多个layer的啊
19  available_blobs->erase(blob_name);
20  bool need_backward = blob_need_backward_[blob_id];
21  // Check if the backpropagation on bottom_id should be skipped
22  if (layer_param.propagate_down_size() > 0) {
23    need_backward = layer_param.propagate_down(bottom_id);
24  }
25  bottom_need_backward_[layer_id].push_back(need_backward);
26  return blob_id;
27}
28

 1
 2// Helper for Net::Init: add a new top blob to the net.
 3template <typename Dtype>
 4void Net<Dtype>::AppendTop(const NetParameter& param, const int layer_id,
 5                           const int top_id, set<string>* available_blobs,
 6                           map<string, int>* blob_name_to_idx) {
 7  shared_ptr<LayerParameter> layer_param(
 8      new LayerParameter(param.layer(layer_id)));
 9  const string& blob_name = (layer_param->top_size() > top_id) ?
10      layer_param->top(top_id) : "(automatic)";
11  // Check if we are doing in-place computation
12  if (blob_name_to_idx && layer_param->bottom_size() > top_id &&
13      blob_name == layer_param->bottom(top_id)) {
14    // In-place computation
15    LOG_IF(INFO, Caffe::root_solver())
16        << layer_param->name() << " -> " << blob_name << " (in-place)";
17    top_vecs_[layer_id].push_back(blobs_[(*blob_name_to_idx)[blob_name]].get());
18    top_id_vecs_[layer_id].push_back((*blob_name_to_idx)[blob_name]);
19  } else if (blob_name_to_idx &&
20             blob_name_to_idx->find(blob_name) != blob_name_to_idx->end()) {
21    // If we are not doing in-place computation but have duplicated blobs,
22    // raise an error.
23    LOG(FATAL) << "Top blob '" << blob_name
24               << "' produced by multiple sources.";
25  } else {
26    // Normal output.
27    if (Caffe::root_solver()) {
28      LOG(INFO) << layer_param->name() << " -> " << blob_name;
29    }
30    shared_ptr<Blob<Dtype> > blob_pointer(new Blob<Dtype>());
31    const int blob_id = blobs_.size();
32    // 　先取blob_id，再push_back 说明blob_id是从0开始递增的整数．
33    //  blob_id都是在appendTop时定义的
34    blobs_.push_back(blob_pointer);
35    blob_names_.push_back(blob_name);
36    blob_need_backward_.push_back(false);
37    if (blob_name_to_idx) { (*blob_name_to_idx)[blob_name] = blob_id; }
38    top_id_vecs_[layer_id].push_back(blob_id);
39    top_vecs_[layer_id].push_back(blob_pointer.get());
40  }
41  // 　维护一个出现过的top　blob集合，目的是判断后面所接layer的bottom blob名称的合法性
42  // 　这里要判断　available_blobs　是否为NULL是因为，可能构造匿名的top blob,此时available_blobs
43  //  和　blob_name_to_idx　都会传NULL 值
44  if (available_blobs) { available_blobs->insert(blob_name); }
45}
46

代码比较好懂，解析部分见注释... 唯一值得注意的地方是available_blobs这部分的维护．在AppendBottom中

1
2  // 　？？？　怎么就erase了，一个top后面可以接多个layer的啊
3  available_blobs->erase(blob_name);
4

每次处理完一个bottom就从available_blobs 去掉．

然而，一个top blob明明是可以输入给多个layer的．．．？？？？？　不明白

查阅资料看到　What does InsertSplit() / SplitLayer do? #767

原来

// Create split layer for any input blobs used by other layers as bottom // blobs more than once.

对于被多个layer用作bottom的top blob,做了分身...

可以参考Net的网络层的构建(源码分析)

caffe 源码学习笔记(3) Net

Net 基本介绍

Net 实现细节

Posts in this Series