Faster-rcnn数据出来流程:
数据读取流程
__main__
----------------combined_roidb
------------get_imdb(factory.py)-------------coco(imdb)----------
------------set_proposal_method--------------gt_roidb(coco.py)--------_load_coco_annotation---------getAnnIds(pycocotools)
|---------loadAnns(pycocotools)
------------get_training_roidb-----------prepare_roidb(roi_data_layer/roidb.py)--------gt_overlaps
--------max_classes
----------------train_net
---------SolverWrapper---------add_bbox_regression_targets(roi_data_layer/roidb.py)计算boundingbox的均值与方差,进行loss计算时需要对box进行归一化
---------train_model
数据输入层:
RoIDataLayer---------------------
forward-----------------
_get_next_minibatch-------------------_get_next_minibatch_inds
-------------------get_minibatch
-------------------_get_image_blob,读图像并放入blob中
数据读取流程
__main__
----------------combined_roidb
------------get_imdb(factory.py)-------------coco(imdb)----------
------------set_proposal_method--------------gt_roidb(coco.py)--------_load_coco_annotation---------getAnnIds(pycocotools)
|---------loadAnns(pycocotools)
------------get_training_roidb-----------prepare_roidb(roi_data_layer/roidb.py)--------gt_overlaps
--------max_classes
----------------train_net
---------SolverWrapper---------add_bbox_regression_targets(roi_data_layer/roidb.py)计算boundingbox的均值与方差,进行loss计算时需要对box进行归一化
---------train_model
数据输入层:
RoIDataLayer---------------------
forward-----------------
_get_next_minibatch-------------------_get_next_minibatch_inds
-------------------get_minibatch
-------------------_get_image_blob,读图像并放入blob中
-------------------将事先准备好的标签放入blob中