Faster RCNN Generate Proposals (#12056)

* Add proposals generation operator for Faster-RCNN.

paddle/fluid/API.spec

@@ -299,6 +299,7 @@ paddle.fluid.layers.ssd_loss ArgSpec(args=['location', 'confidence', 'gt_box', '
 paddle.fluid.layers.detection_map ArgSpec(args=['detect_res', 'label', 'class_num', 'background_label', 'overlap_threshold', 'evaluate_difficult', 'has_state', 'input_states', 'out_states', 'ap_version'], varargs=None, keywords=None, defaults=(0, 0.3, True, None, None, None, 'integral'))
 paddle.fluid.layers.rpn_target_assign ArgSpec(args=['loc', 'scores', 'anchor_box', 'gt_box', 'rpn_batch_size_per_im', 'fg_fraction', 'rpn_positive_overlap', 'rpn_negative_overlap'], varargs=None, keywords=None, defaults=(256, 0.25, 0.7, 0.3))
 paddle.fluid.layers.anchor_generator ArgSpec(args=['input', 'anchor_sizes', 'aspect_ratios', 'variance', 'stride', 'offset', 'name'], varargs=None, keywords=None, defaults=(None, None, [0.1, 0.1, 0.2, 0.2], None, 0.5, None))
+paddle.fluid.layers.generate_proposals ArgSpec(args=['scores', 'bbox_deltas', 'im_info', 'anchors', 'variances', 'pre_nms_top_n', 'post_nms_top_n', 'nms_thresh', 'min_size', 'eta', 'name'], varargs=None, keywords=None, defaults=(6000, 1000, 0.5, 0.1, 1.0, None))
 paddle.fluid.layers.iou_similarity ArgSpec(args=[], varargs='args', keywords='kwargs', defaults=None)
 paddle.fluid.layers.box_coder ArgSpec(args=[], varargs='args', keywords='kwargs', defaults=None)
 paddle.fluid.layers.polygon_box_transform ArgSpec(args=[], varargs='args', keywords='kwargs', defaults=None)

paddle/fluid/operators/detection/CMakeLists.txt

@@ -29,6 +29,6 @@ target_assign_op.cu)
 detection_library(polygon_box_transform_op SRCS polygon_box_transform_op.cc
 polygon_box_transform_op.cu)
 detection_library(rpn_target_assign_op SRCS rpn_target_assign_op.cc)
-
-# Export local libraries to parent
+detection_library(generate_proposals_op SRCS generate_proposals_op.cc)
+#Export local libraries to parent
 set(DETECTION_LIBRARY ${LOCAL_DETECTION_LIBS} PARENT_SCOPE)

python/paddle/fluid/layers/detection.py

@@ -39,6 +39,7 @@ __all__ = [
     'detection_map',
     'rpn_target_assign',
     'anchor_generator',
+    'generate_proposals',
 ]
 
 __auto__ = [
@@ -1253,3 +1254,73 @@ def anchor_generator(input,
     anchor.stop_gradient = True
     var.stop_gradient = True
     return anchor, var
+
+
+def generate_proposals(scores,
+                       bbox_deltas,
+                       im_info,
+                       anchors,
+                       variances,
+                       pre_nms_top_n=6000,
+                       post_nms_top_n=1000,
+                       nms_thresh=0.5,
+                       min_size=0.1,
+                       eta=1.0,
+                       name=None):
+    """
+    ** Generate proposal labels Faster-RCNN **
+	
+	This operation proposes RoIs according to each box with their probability to be a foreground object and 
+	the box can be calculated by anchors. Bbox_deltais and scores to be an object are the output of RPN. Final proposals
+	could be used to train detection net.
+
+	For generating proposals, this operation performs following steps:
+
+	1. Transposes and resizes scores and bbox_deltas in size of (H*W*A, 1) and (H*W*A, 4)
+ 	2. Calculate box locations as proposals candidates. 
+	3. Clip boxes to image
+	4. Remove predicted boxes with small area. 
+	5. Apply NMS to get final proposals as output.
+	
+      
+	Args:
+		scores(Variable): A 4-D Tensor with shape [N, A, H, W] represents the probability for each box to be an object.
+			N is batch size, A is number of anchors, H and W are height and width of the feature map.
+		bbox_deltas(Variable): A 4-D Tensor with shape [N, 4*A, H, W] represents the differece between predicted box locatoin and anchor location. 
+		im_info(Variable): A 2-D Tensor with shape [N, 3] represents origin image information for N batch. Info contains height, width and scale
+			between origin image size and the size of feature map.
+		anchors(Variable):   A 4-D Tensor represents the anchors with a layout of [H, W, A, 4]. H and W are height and width of the feature map,
+              		num_anchors is the box count of each position. Each anchor is in (xmin, ymin, xmax, ymax) format an unnormalized.
+		variances(Variable): The expanded variances of anchors with a layout of [H, W, num_priors, 4]. Each variance is in (xcenter, ycenter, w, h) format.
+		pre_nms_top_n(float): Number of total bboxes to be kept per image before NMS. 6000 by default.
+		post_nms_top_n(float): Number of total bboxes to be kept per image after NMS. 1000 by default.
+		nms_thresh(float): Threshold in NMS, 0.5 by default.
+		min_size(float): Remove predicted boxes with either height or width < min_size. 0.1 by default.
+		eta(float): Apply in adaptive NMS, if adaptive threshold > 0.5, adaptive_threshold = adaptive_threshold * eta in each iteration.
+    """
+    helper = LayerHelper('generate_proposals', **locals())
+
+    rpn_rois = helper.create_tmp_variable(dtype=bbox_deltas.dtype)
+    rpn_roi_probs = helper.create_tmp_variable(dtype=scores.dtype)
+    helper.append_op(
+        type="generate_proposals",
+        inputs={
+            'Scores': scores,
+            'BboxDeltas': bbox_deltas,
+            'ImInfo': im_info,
+            'Anchors': anchors,
+            'Variances': variances
+        },
+        attrs={
+            'pre_nms_topN': pre_nms_top_n,
+            'post_nms_topN': post_nms_top_n,
+            'nms_thresh': nms_thresh,
+            'min_size': min_size,
+            'eta': eta
+        },
+        outputs={'RpnRois': rpn_rois,
+                 'RpnRoiProbs': rpn_roi_probs})
+    rpn_rois.stop_gradient = True
+    rpn_roi_probs.stop_gradient = True
+
+    return rpn_rois, rpn_roi_probs

python/paddle/fluid/tests/test_detection.py

@@ -201,5 +201,44 @@ class TestDetectionMAP(unittest.TestCase):
         print(str(program))
 
 
+class TestGenerateProposals(unittest.TestCase):
+    def test_generate_proposals(self):
+        data_shape = [20, 64, 64]
+        images = fluid.layers.data(
+            name='images', shape=data_shape, dtype='float32')
+        im_info = fluid.layers.data(
+            name='im_info', shape=[1, 3], dtype='float32')
+        anchors, variances = fluid.layers.anchor_generator(
+            name='anchor_generator',
+            input=images,
+            anchor_sizes=[32, 64],
+            aspect_ratios=[1.0],
+            variance=[0.1, 0.1, 0.2, 0.2],
+            stride=[16.0, 16.0],
+            offset=0.5)
+        num_anchors = anchors.shape[2]
+        scores = fluid.layers.data(
+            name='scores', shape=[1, num_anchors, 8, 8], dtype='float32')
+        bbox_deltas = fluid.layers.data(
+            name='bbox_deltas',
+            shape=[1, num_anchors * 4, 8, 8],
+            dtype='float32')
+        rpn_rois, rpn_roi_probs = fluid.layers.generate_proposals(
+            name='generate_proposals',
+            scores=scores,
+            bbox_deltas=bbox_deltas,
+            im_info=im_info,
+            anchors=anchors,
+            variances=variances,
+            pre_nms_top_n=6000,
+            post_nms_top_n=1000,
+            nms_thresh=0.5,
+            min_size=0.1,
+            eta=1.0)
+        self.assertIsNotNone(rpn_rois)
+        self.assertIsNotNone(rpn_roi_probs)
+        print(rpn_rois.shape)
+
+
 if __name__ == '__main__':
     unittest.main()