def match_targets_to_proposals(self, proposal, target):
box1, box2 = proposal.bbox, target.bbox
box1_np = box1.data.cpu().numpy()
box2_np = box2.data.cpu().numpy()
ch_box1 = box1_np.copy()
ch_box11 = ch_box1[:, [0, 1, 3, 4, 6]]
ch_box2 = box2_np.copy()
ch_box22 = ch_box2[:, [0, 1, 3, 4, 6]]
match_quality_matrix2 = box_np_ops.riou_cc(ch_box22, ch_box11)
match_quality_matrix2 = torch.from_numpy(
match_quality_matrix2).float().to(box1.device)
#match_quality_matrix = boxlist_iou(proposal, target)
#print(match_quality_matrix)
#print(match_quality_matrix2)
matched_idxs = self.proposal_matcher(match_quality_matrix2)
# print('matched_idxs', matched_idxs)
# Fast RCNN only need "labels" field for selecting the targets
target = target.copy_with_fields("labels")
#print("this is target")
#print(len(target) , target)
# get the targets corresponding GT for each proposal
# NB: need to clamp the indices because we can have a single
# GT in the image, and matched_idxs can be -2, which goes
# out of bounds
matched_targets = target[matched_idxs.clamp(min=0)]
matched_targets.add_field("matched_idxs", matched_idxs)
return matched_targets
def _compare(self, boxes1, boxes2):
"""Compute pairwise IOU similarity between the two BoxLists.
Args:
boxlist1: BoxList holding N boxes.
boxlist2: BoxList holding M boxes.
Returns:
A tensor with shape [N, M] representing pairwise iou scores.
"""
return box_np_ops.riou_cc(boxes1, boxes2)
def bev_box_overlap(boxes, qboxes, criterion=-1, stable=True):
if USE_GPU:
riou = rotate_iou_gpu_eval(boxes, qboxes, criterion)
else:
riou = box_np_ops.riou_cc(boxes, qboxes)
return riou
def bev_box_overlap(boxes, qboxes, criterion=-1, stable=True):
riou = box_np_ops.riou_cc(boxes, qboxes)
return riou
#loading data anti to clock-wise
rn = ((boxes[n, 0], boxes[n,
1]), (boxes[n,
2], boxes[n,
3]), -boxes[n, 4])
rk = ((query_boxes[k, 0], query_boxes[k, 1]),
(query_boxes[k, 2], query_boxes[k, 3]), -query_boxes[k, 4])
int_pts = cv2.rotatedRectangleIntersection(rk, rn)[1]
if int_pts is not None:
order_pts = cv2.convexHull(int_pts, returnPoints=True)
int_area = cv2.contourArea(order_pts)
overlaps[n, k] = int_area * 1.0 / (query_area + box_area -
int_area)
return overlaps
# (x,y,w,l,r)
box1 = np.array([[0, 0, 2, 2, 0.3]], dtype=np.float32)
box2 = np.array([[2, 0, 4, 1, 0.3]], dtype=np.float32)
iou1 = rotate_iou_gpu(box1, box2) # 1/7
print(iou1)
iou2 = riou_cc(box1, box2) # 1/7
print(iou2)
box1[..., [0, 1]] += 100
box2[..., [0, 1]] += 100
iou3 = cv2_rbbx_overlaps(box1, box2)
print(iou3)
def bev_box_overlap(boxes, qboxes, criterion=-1, stable=False):
if stable:
riou = box_np_ops.riou_cc(boxes, qboxes)
else:
riou = rotate_iou_gpu_eval(boxes, qboxes, criterion)
return riou
