Crate caffe2op_collect

source ·

Structs

  • | Merge RPN proposals generated at multiple | FPN levels and then distribute those | proposals to their appropriate FPN | levels for Faster RCNN. | | An anchor at one FPN level may predict | an RoI that will map to another level, | hence the need to redistribute the proposals. | | Only inference is supported. To train, | please use the original Python operator | in Detectron. | | Inputs and outputs are examples only; | if min/max levels change, the number | of inputs and outputs, as well as their | level numbering, will change. | | C++ implementation of | | CollectAndDistributeFpnRpnProposalsOp | Merge RPN proposals generated at multiple | FPN levels and then distribute those | proposals to their appropriate FPN | levels for Faster RCNN. An anchor at | one FPN level may predict an RoI that | will map to another level, hence the | need to redistribute the proposals. | | Reference: facebookresearch/Detectron/detectron/ops/collect_and_distribute_fpn_rpn_proposals.py |

Functions

  • | Updates arr to be indices that would | sort the array. Implementation of https://docs.scipy.org/doc/numpy/reference/generated/numpy.argsort.html |
  • Compute the area of an array of boxes.
  • | Determine which FPN level each RoI in | a set of RoIs should map to based on the | heuristic in the FPN paper. |
  • | Update out_filtered and out_indices | with rows from rois where lvl matches | value in lvls passed in. |
  • | Sort RoIs from highest to lowest individual | RoI score based on values from scores | array and limit to n results |