R-CNN

dragon.operators.contrib.rcnn.ops.Proposal(
   inputs,
   strides,
   ratios,
   scales,
   pre_nms_top_n=6000,
   post_nms_top_n=300,
   nms_thresh=0.7,
   min_size=16,
   min_level=2,
   max_level=5,
   canonical_scale=224,
   canonical_level=4,
   **kwargs
)

Generate Regional Proposals, introduced by [Ren et.al, 2015].

Multi-Level proposals was introduced by [Lin et.al, 2017].

For single level proposals(e.g. C4), the inputs should be: [cls_probs, bbox_deltas, im_info].

For multiple level proposals(e.g. FPN), the inputs should be: [cls_score/Px, …] + [cls_probs, bbox_deltas, im_info].

Parameters:
  • inputs (sequence of Tensor) – The inputs.
  • strides (sequence of int) – The strides of anchors.
  • ratios (sequence of float) – The ratios of anchors.
  • scales (sequence of float) – The scales of anchors.
  • pre_nms_top_n (int, optional, default=6000) – The number of anchors before nms.
  • post_nms_top_n (int, optional, default=300) – The number of anchors after nms.
  • nms_thresh (float, optional, default=0.7) – The threshold of nms.
  • min_size (int, optional, default=16) – The min size of anchors.
  • min_level (int, optional, default=2) – Finest level of the FPN pyramid.
  • max_level (int, optional, default=5) – Coarsest level of the FPN pyramid.
  • canonical_scale (int, optional, default=224) – The baseline scale of mapping policy.
  • canonical_level (int, optional, default=4) – Heuristic level of the canonical scale.
Returns:

The proposals.

Return type:

Tensor