Faster RCNN minibatch.py解读

最新推荐文章于 2020-06-17 10:44:26 发布

原创

最新推荐文章于 2020-06-17 10:44:26 发布 · 3.8k 阅读

9 ·

CC 4.0 BY-SA版权

文章标签：

#faster-r-cnn

minibatch.py 的功能是： Compute minibatch blobs for training a Fast R-CNN network. 与roidb不同的是， minibatch中存储的并不是完整的整张图像图像，而是从图像经过转换后得到的四维blob以及从图像中截取的proposals，以及与之对应的labels等

在整个faster rcnn训练中，有两处用到了minibatch.py，一处是rpn的开始数据输入，另一处自然是fast rcnn的数据输入。分别见stage1_rpn_train.pt与stage1_fast_rcnn_train.py的最前面，如下：
stage1_rpn_train.pt：

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': 21"
  }
}

stage1_fast_rcnn_train.py：

name: "VGG_CNN_M_1024"
layer {
  name: 'data'
  type: 'Python'
  top: 'data'
  top: 'rois'
  top: 'labels'
  top: 'bbox_targets'
  top: 'bbox_inside_weights'
  top: 'bbox_outside_weights'
  python_param {
    module: 'roi_data_layer.layer'
    layer: 'RoIDataLayer'
    param_str: "'num_classes': 21"
  }
}

如上，共同的数据定义层为roi_data_layer.layer，在layer.py中，观察前向传播：

 def forward(self, bottom, top):
        """Get blobs and copy them into this layer's top blob vector."""
        blobs = self._get_next_minibatch()

        for blob_name, blob in blobs.iteritems():
            top_ind = self._name_to_top_map[blob_name]
            # Reshape net's input blobs
            top[top_ind].reshape(*(blob.shape))
            # Copy data into net's input blobs
            top[top_ind].data[...] = blob.astype(np.float32, copy=False)
 def _get_next_minibatch(self):
        """Return the blobs to be used for the next minibatch.

        If cfg.TRAIN.USE_PREFETCH is True, then blobs will be computed in a
        separate process and made available through self._blob_queue.
        """
        if cfg.TRAIN.USE_PREFETCH:
            return self._blob_queue.get()
        else:
            db_inds = self._get_next_minibatch_inds()
            minibatch_db = [self._roidb[i] for i in db_inds]
            return get_minibatch(minibatch_db, self._num_classes)