tensorflow tf.py_func

tf.py_func

See this function in the tensorflow implementation of faster rcnn

1	rois,rpn_scores = tf.py_func(proposal_layer,[rpn_cls_prob,rpn_bbox_pred, self ._im_info, self .mode, self ._feat_stride, self ._anchors, self ._num_anchors],[tf.float32,tf.float32],name = "proposal" )

　　Explanation on the tensorflow official website

py_func(
    func,     inp,     Tout,     stateful=True,     name=None )
Wrap a python function as a tensorflow operator
The python function proposal_layer takes numpy matrices as input and output, making the function an operator in the tensorflow graph
Define a simple sinh function in the tensorflow graph:

def my_func(x):

  
# x will be a numpy array with the contents of the placeholder below

  
  return np.sinh(x)

inp =tf.placeholder(tf.float32)

y =tf.py_func(my_func, [inp], tf.float32)

When tf.py_func defines a multi-output function, the output variable type needs to be framed with [ ];

When tf.py_func defines a single output function, the output variable type can no longer be framed with [ ];

This needs everyone's attention!

E.g:

multivariate

[python] view plain copy

1. def _proposal_layer(self, rpn_cls_prob, rpn_bbox_pred, name):  
2.   with tf.variable_scope(name) as scope:  
3.     rois, rpn_scores, inds= tf.py_func(proposal_layer,  
4.                                   [rpn_cls_prob, rpn_bbox_pred, self._im_info, self._mode,  
5.                                    self._feat_stride, self._anchors, self._num_anchors],  
6.                                   [tf.float32, tf.float32,tf.int64])  
7.     # rois.set_shape([None, 5])  
8.     # rpn_scores.set_shape([None, 1])  
9.   rois.set_shape([1,None,None,self._num_anchors*5])  
10.   rpn_scores.set_shape([1,None,None,self._num_anchors*1])  
11.   return rois, rpn_scores,inds  

When univariate

[python] view plain copy

1. def _draw_proposals_to_image(self,rois,scores,inds,keep_inds,stride,name):  
2.   with tf.variable_scope(name) as scope:  
3.     mask = tf.py_func(  
4.       proposals_to_image,  
5.       [rois, scores, inds, keep_inds,stride],  
6.       tf.float32)  
7.   
8.     mask = tf.stop_gradient(mask)  
9.     mask.set_shape([1, None, None, cfg.TRAIN.BATCH_SIZE])  
10.   
11.     return mask