faster-rcnn代码阅读2

二、训练

接下来回到train.py第160行，通过调用sw.train_model方法进行训练：

     def train_model(self, max_iters):
         """Network training loop."""
         last_snapshot_iter = -1
         timer = Timer()
         model_paths = []
         while self.solver.iter < max_iters:
             # Make one SGD update
             timer.tic()
             self.solver.step(1)
             timer.toc()
             if self.solver.iter % (10 * self.solver_param.display) == 0:
                 print 'speed: {:.3f}s / iter'.format(timer.average_time)

             if self.solver.iter % cfg.TRAIN.SNAPSHOT_ITERS == 0:
                 last_snapshot_iter = self.solver.iter
                 model_paths.append(self.snapshot())

         if last_snapshot_iter != self.solver.iter:
             model_paths.append(self.snapshot())
         return model_paths

方法中的self.solver.step(1)即是网络进行一次前向传播和反向传播。前向传播时，数据流会从第一层流动到最后一层，最后计算出loss，然后loss相对于各层输入的梯度会从最后一层计算回第一层。下面逐层来介绍faster-rcnn算法的运行过程。

2.1、input-data layer

第一层是由python代码构成的，其prototxt描述为：

layer {
  name: 'input-data'
  type: 'Python'
  top: 'data'
  top: 'im_info'
  top: 'gt_boxes'
  python_param {
    module: 'roi_data_layer.layer'
    layer: 'RoIDataLayer'
    param_str: "'num_classes': 2"
  }
}

从中可以看出，input-data层有三个输出：data、im_info、gt_boxes，其实现为RoIDataLayer类。这一层对数据的预处理操作为：对图片进行长宽等比例缩放，使短边缩放至600；如果缩放后，长边的长度大于1000，则以长边为基准，将长边缩放至1000，短边作相应的等比例缩放。这一层的3个输出分别为：

1、data：1, 3, h, w(一个batch只支持输入一张图)

2、im_info: im_info[0], im_info[1], im_info[2]分别为h, w, target_size/im_origin_size(缩放比例)

3、gt_boxes: (x1, y1, x2, y2, cls)

预处理部分涉及到的函数有_get_next_minibatch，get_minibatch，_get_image_blob，prep_im_for_blob，im_list_to_blob。

网络在构造过程中（即self.solver = caffe.SGDSolver(solver_prototxt)）会调用该类的setup方法：

 __C.TRAIN.IMS_PER_BATCH = 1
 __C.TRAIN.SCALES = [600]
 __C.TRAIN.MAX_SIZE = 1000
 __C.TRAIN.HAS_RPN = True
 __C.TRAIN.BBOX_REG = True

     def setup(self, bottom, top):
         """Setup the RoIDataLayer."""

         # parse the layer parameter string, which must be valid YAML
         layer_params = yaml.load(self.param_str_)

         self._num_classes = layer_params['num_classes']

         self._name_to_top_map = {}

         # data blob: holds a batch of N images, each with 3 channels
         idx = 0
         top[idx].reshape(cfg.TRAIN.IMS_PER_BATCH, 3,
             max(cfg.TRAIN.SCALES), cfg.TRAIN.MAX_SIZE)
         self._name_to_top_map['data'] = idx
         idx += 1

         if cfg.TRAIN.HAS_RPN:
             top[idx].reshape(1, 3)
             self._name_to_top_map['im_info'] = idx
             idx += 1

             top[idx].reshape(1, 4)
             self._name_to_top_map['gt_boxes'] = idx
             idx += 1
         else: # not using RPN
             # rois blob: holds R regions of interest, each is a 5-tuple
             # (n, x1, y1, x2, y2) specifying an image batch index n and a
             # rectangle (x1, y1, x2, y2)
             top[idx].reshape(1, 5)
             self._name_to_top_map['rois'] = idx
             idx += 1

             # labels blob: R categorical labels in [0, ..., K] for K foreground
             # classes plus background
             top[idx].reshape(1)
             self._name_to_top_map['labels'] = idx
             idx += 1

             if cfg.TRAIN.BBOX_REG:
                 # bbox_targets blob: R bounding-box regression targets with 4
                 # targets per class
                 top[idx].reshape(1, self._num_classes * 4)
                 self._name_to_top_map['bbox_targets'] = idx
                 idx += 1

                 # bbox_inside_weights blob: At most 4 targets per roi are active;
                 # thisbinary vector sepcifies the subset of active targets
                 top[idx].reshape(1, self._num_classes * 4)
                 self._name_to_top_map['bbox_inside_weights'] = idx
                 idx += 1

                 top[idx].reshape(1, self._num_classes * 4)
                 self._name_to_top_map['bbox_outside_weights'] = idx
                 idx += 1

         print 'RoiDataLayer: name_to_top:', self._name_to_top_map
         assert len(top) == len(self._name_to_top_map)

主要是对输出的shape进行定义。要说明的是，在前向传播的过程中，仍然会对输出的各top的shape进行重定义，并且二者定义的shape往往都是不同的。