def nms_gpu(boxes, scores, thresh):
"""
:param boxes: (N, 5) [x1, y1, x2, y2, ry]
:param scores: (N)
:param thresh:
:return:
"""
# areas = (x2 - x1) * (y2 - y1)
# print("scores :", scores)
# scores : tensor([[ -1.9006, 2.7015, -210.2983, -205.6284],
# [ -1.9006, 2.7015, -210.2983, -205.6284],
# [ -1.9006, 2.7015, -210.2983, -205.6284],... => sane value printed = problem!
order = scores.sort(0, descending=True)[1]
# print("order :", order)
# order : tensor([[31, 31, 31, 31],
# [30, 30, 30, 30],
# [28, 28, 28, 28],...
boxes = boxes[order].contiguous()
# print("boxes :", boxes)
# boxes : tensor([[[ 3.2608, 5.4325, 4.3231, 6.0319, 1.5068],
# [ 3.2608, 5.4325, 4.3231, 6.0319, 1.5068],
# [ 3.2608, 5.4325, 4.3231, 6.0319, 1.5068],...
keep = torch.LongTensor(boxes.size(0))
# print("keep", keep)
num_out = iou3d_cuda.nms_gpu(boxes, keep, thresh)
# print("num_out :",num_out) # 6
return order[keep[:num_out].cuda()].contiguous()
def nms_gpu(boxes, scores, thresh):
"""
:param boxes: (N, 5) [x1, y1, x2, y2, ry]
:param scores: (N)
:param thresh:
:return:
"""
# areas = (x2 - x1) * (y2 - y1)
order = scores.sort(0, descending=True)[1]
boxes = boxes[order].contiguous()
keep = torch.LongTensor(boxes.size(0))
num_out = iou3d_cuda.nms_gpu(boxes, keep, thresh)
return order[keep[:num_out].cuda()].contiguous()
def nms_gpu(boxes, scores, thresh, box_mode='wlh'):
"""
filter overlapped boxes based on bev iou
:param boxes(torch): (N, 7), [x, y, z, h, w, l, ry], torch tensor with type float32
:param scores: (N)
:param thresh: in the range [0, 1] for iou thresh
:return:
"""
# areas = (x2 - x1) * (y2 - y1)
boxes = utils.boxes3d_to_bev_torch(boxes, box_mode, rect=True)
order = scores.sort(0, descending=True)[1]
boxes = boxes[order].contiguous()
keep = torch.LongTensor(boxes.size(0))
num_out = iou3d_cuda.nms_gpu(boxes, keep, thresh)
return order[keep[:num_out].cuda()].contiguous()