class_num =8
path ='./ground.txt'
processed_path ='./Data/processed/'
with codecs.open(path, 'r', 'utf-8')as f:
lines = f.readlines()
for num, image_idxin enumerate(lines):
"""print(num+1, ' ', image_idx, '\n')"""
ground_truth_dic = load_annotation(image_idx)
image_path = ('./Data/Images/' + image_idx).strip('\n') +'.jpg'
# print(image_path)
img = cv2.imread(image_path)
# print(img.shape)
img_lbl, regions = selectivesearch.selective_search(img, scale=1000, sigma=0.9, min_size=1000)
labels = []
for rin regions:
x, y, w, h = r['rect']# x,y,w,h中，(x,y)是bounding box的左下角的坐標，w,h代表寬和高
proposal_vertice = [x +1, y, x + w, y + h, w, h]
proposal_bbox = [x, y, (x + w -1), (y + h -1)]
label = np.zeros(class_num *5 -4, dtype=np.float32)
iou_val =0
# 字典的items()函數：返回可遍歷的(鍵, 值) 元組數組
for ground_truth, class_idxin ground_truth_dic.items():
# ground_truth = list(ground_truth)
xmin = (2 * ground_truth[0] - ground_truth[2]) /2.0
ymin = (2 * ground_truth[1] - ground_truth[3]) /2.0
ground_truth = [xmin, ymin, ground_truth[2], ground_truth[3]]
iou_val = IOU(ground_truth, proposal_bbox)
"""
(px,py,pw,ph)是selectivesearch生成的bounding box的中心點坐標、寬、高
（gx,gy,gw,gh)是解析xml信息中標註的groundtruth中的標註框的中心點坐標、寬、高
"""
px =float(proposal_vertice[0]) +float(proposal_vertice[4] /2.0)# 中心點X
py =float(proposal_vertice[1]) +float(proposal_vertice[5] /2.0)# 中心點Y
pw =float(proposal_vertice[4])# w
ph =float(proposal_vertice[5])# h
gx =float(ground_truth[0])# 中心點X
gy =float(ground_truth[1])# 中心點Y
gw =float(ground_truth[2])# W
gh =float(ground_truth[3])# H
if iou_val <0.7:
label[0] =1
elif iou_val >0.7:
label[0] =0
label[class_idx] =1
# 下面這行是啥意思，不懂
label[class_num + (class_idx -1) *4: class_num + (class_idx -1) *4 +4] = \
[((gx - px) / pw), ((gy - py) / ph), (np.log(gw / pw)), (np.log(gh / ph))]
break
for iin range(len(proposal_bbox)):
proposal_bbox[i] = (proposal_bbox[i] /16.0)
proposal_bbox.insert(0, 0)
proposal_bbox.insert(0, iou_val)
proposal_bbox.extend(label)
labels.append(proposal_bbox)
"""
for k in labels:
print(k, '\n')
"""
view_bar("Process image of %s" % image_path, num +1, len(lines))
if True:
if not os.path.exists(processed_path):? os.makedirs(processed_path)
np.save((os.path.join(processed_path, image_idx.split('.')[0].strip())
+'_data.npy'), labels)
最終每壹張圖片生成壹個.npy文件。對每個圖片生成的候選框進行選擇，通過選擇的候選框將生成相應的labels,壹個圖片中所有的labels生成壹個.npy文件。