easy_vision.python.model.rfcn

easy_vision.python.model.rfcn.fcn_head

class easy_vision.python.model.rfcn.fcn_head.FCNHead(feature_dict, head_config, label_dict=None, mode='predict')

Bases: easy_vision.python.model.faster_rcnn.rcnn_head.RCNNHead

for the second stage of rfcn: classification based on position sensitive pooling

__init__(feature_dict, head_config, label_dict=None, mode='predict')

see rcnn_head.py for parameter explanations

build_predict_graph()

input: proposal_boxes, feature_map output:

refined_box_encodings_with_background, class_predictions_with_background

steps:

1. classify block: extract classification features with backbone features as input.
2. using box_predictor to generate box scores and encodings predictions. 2.1 conv2d to generate position sensitive score maps and box offset maps. 2.2 psroipooling on score preds and box_encodings.

easy_vision.python.model.rfcn.rfcn

class easy_vision.python.model.rfcn.rfcn.RFCN(model_config, feature_dict, label_dict=None, mode='predict', categories=None)

Bases: easy_vision.python.model.faster_rcnn.faster_rcnn.FasterRcnn

__init__(model_config, feature_dict, label_dict=None, mode='predict', categories=None)

x.__init__(…) initializes x; see help(type(x)) for signature

build_loss_graph()

build_predict_graph()

classmethod create_class(name)

get_scopes_of_levels()

return a list of variable scope list order by levels ( outputs -> heads -> backbone -> inputs).