how/where the data blob is created in py faster rcnn?

[dusa]

Hi,

Can anyone let me know how the data blob is created in py faster rcnn?? I have been checking all the train code, factory, imdb and pascal_voc under dataset and the layer. I wish to add another set of images (you could assume depth, so it is a pair of the input image, should have same order). 

I see that the data is defined in prototxt as :

name: "ZF"

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"

  }

}

so I need another top for the new image set, and I see that in the layer.py (under roi data layer) the blob is accessed and reshaped as it is defined in prototxt.

  # 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

Since idx is a counter, I am assuming the data blob is read in order. 

 fine, but how does the code prepare the data blob? I can not find it exactly. Thanks!!!!