def traditional_nms(self,
boxes,
masks,
scores,
iou_threshold=0.5,
conf_thresh=0.05):
import pyximport
pyximport.install(setup_args={"include_dirs": np.get_include()},
reload_support=True)
from utils.cython_nms import nms as cnms
num_classes = scores.size(0)
idx_lst = []
cls_lst = []
scr_lst = []
# Multiplying by max_size is necessary because of how cnms computes its area and intersections
boxes = boxes * cfg.max_size
for _cls in range(num_classes):
cls_scores = scores[_cls, :]
# threshold 넘는 confidence score 남김
conf_mask = cls_scores > conf_thresh
# 점수 내림차순으로 정렬
idx = torch.arange(cls_scores.size(0), device=boxes.device)
cls_scores = cls_scores[conf_mask]
idx = idx[conf_mask]
# 없으면 건너뛰기
if cls_scores.size(0) == 0:
continue
preds = torch.cat([boxes[conf_mask], cls_scores[:, None]],
dim=1).cpu().numpy()
# 예측에 대해 iou 계산 후 threshold 넘는 것 제거
keep = cnms(preds, iou_threshold)
keep = torch.Tensor(keep, device=boxes.device).long()
idx_lst.append(idx[keep])
cls_lst.append(keep * 0 + _cls)
scr_lst.append(cls_scores[keep]) # 남은 것들 list에 추가
idx = torch.cat(idx_lst, dim=0)
classes = torch.cat(cls_lst, dim=0)
scores = torch.cat(scr_lst, dim=0)
scores, idx2 = scores.sort(0, descending=True)
idx2 = idx2[:cfg.max_num_detections]
scores = scores[:cfg.max_num_detections]
idx = idx[idx2]
classes = classes[idx2]
# Undo the multiplication above
return boxes[idx] / cfg.max_size, masks[idx], classes, scores
def traditional_nms(self,
boxes,
masks,
scores,
iou_threshold=0.5,
conf_thresh=0.05):
import pyximport
pyximport.install(setup_args={"include_dirs": np.get_include()},
reload_support=True)
from utils.cython_nms import nms as cnms
num_classes = scores.shape[0]
idx_lst = []
cls_lst = []
scr_lst = []
# Multiplying by max_size is necessary because of how cnms computes its area and intersections
boxes = boxes * cfg.max_size
for _cls in range(num_classes):
cls_scores = scores[_cls]
conf_mask = cls_scores > conf_thresh
idx = jt.arange(cls_scores.shape[0], device=boxes.device)
cls_scores = cls_scores[conf_mask]
idx = idx[conf_mask]
if cls_scores.shape[0] == 0:
continue
preds = jt.contrib.concat([boxes[conf_mask], cls_scores[:, None]],
dim=1).numpy()
keep = cnms(preds, iou_threshold)
keep = jt.array(keep).int64()
idx_lst.append(idx[keep])
cls_lst.append(keep * 0 + _cls)
scr_lst.append(cls_scores[keep])
idx = jt.contrib.concat(idx_lst, dim=0)
classes = jt.contrib.concat(cls_lst, dim=0)
scores = jt.contrib.concat(scr_lst, dim=0)
scores, idx2 = scores.sort(0, descending=True)
idx2 = idx2[:cfg.max_num_detections]
scores = scores[:cfg.max_num_detections]
idx = idx[idx2]
classes = classes[idx2]
# Undo the multiplication above
return boxes[idx] / cfg.max_size, masks[idx], classes, scores
def traditional_nms_ours(self,
boxes,
masks,
scores,
iou_threshold=0.5,
conf_thresh=0.05):
import pyximport
pyximport.install(setup_args={"include_dirs": np.get_include()},
reload_support=True)
from utils.cython_nms import nms as cnms
num_classes = scores.size(0)
idx_lst = []
cls_lst = []
scr_lst = []
box_lst = []
mask_lst = []
# Multiplying by max_size is necessary because of how cnms computes its area and intersections
boxes = boxes * cfg.max_size
for _cls in range(num_classes):
cls_scores = scores[_cls, :]
_, id = cls_scores.sort(0, descending=True)
id = id[:200].contiguous()
cls_scores = cls_scores[id]
idx = torch.arange(cls_scores.size(0), device=boxes.device)
if cls_scores.size(0) == 0:
continue
preds = torch.cat([boxes[id], cls_scores[:, None]],
dim=1).cpu().numpy()
keep = cnms(preds, iou_threshold)
keep = torch.Tensor(keep, device=boxes.device).long()
m = (cls_scores[keep] > 0.01)
idx_lst.append(idx[keep][m])
cls_lst.append(keep[m] * 0 + _cls)
scr_lst.append(cls_scores[keep][m])
box_lst.append(boxes[id][keep][m])
mask_lst.append(masks[id][keep][m])
idx = torch.cat(idx_lst, dim=0)
classes = torch.cat(cls_lst, dim=0)
scores = torch.cat(scr_lst, dim=0)
boxes = torch.cat(box_lst, dim=0)
masks = torch.cat(mask_lst, dim=0)
scores, idx2 = scores.sort(0, descending=True)
idx2 = idx2[:cfg.max_num_detections]
scores = scores[:cfg.max_num_detections]
classes = classes[idx2]
# Undo the multiplication above
return boxes[idx2] / cfg.max_size, masks[idx2], classes, scores
def traditional_nms(self,
boxes,
masks,
track,
scores,
iou_threshold=0.5,
conf_thresh=0.05):
num_classes = scores.size(0)
idx_lst = []
cls_lst = []
scr_lst = []
# Multiplying by max_size is necessary because of how cnms computes its area and intersections
boxes = boxes * cfg.max_size
for _cls in range(num_classes):
cls_scores = scores[_cls, :]
conf_mask = cls_scores > conf_thresh
idx = torch.arange(cls_scores.size(0), device=boxes.device)
cls_scores = cls_scores[conf_mask]
idx = idx[conf_mask]
if cls_scores.size(0) == 0:
continue
preds = torch.cat([boxes[conf_mask], cls_scores[:, None]],
dim=1).cpu().numpy()
keep = cnms(preds, iou_threshold)
keep = torch.Tensor(keep, device=boxes.device).long()
idx_lst.append(idx[keep])
cls_lst.append(keep * 0 + _cls)
scr_lst.append(cls_scores[keep])
idx = torch.cat(idx_lst, dim=0)
classes = torch.cat(cls_lst, dim=0)
scores = torch.cat(scr_lst, dim=0)
scores, idx2 = scores.sort(0, descending=True)
idx2 = idx2[:cfg.max_num_detections]
scores = scores[:cfg.max_num_detections]
idx = idx[idx2]
classes = classes[idx2]
# Undo the multiplication above
return boxes[idx] / cfg.max_size, masks[idx], track[
idx], classes, scores
