def batch_analysis_c6(meta_file, cfg_file, wgt_file, thresh, nms, img_path,
xml_path):
image_list = listdir(img_path, '.jpg')
image_num = len(image_list)
meta = dn.load_meta(meta_file)
net = dn.load_net(cfg_file, wgt_file, 0)
# meta_fr = dn.load_meta(meta_file_fr)
# net_fr = dn.load_net(cfg_file_fr,wgt_file_fr,0)
move_count = 0
for j, image_path in enumerate(image_list):
print(str(j) + '/' + str(image_num) + " " + image_path)
image_name = getFileName(image_path)
img_save_path = os.path.join(img_path, image_name + '.jpg')
xml_save_path = os.path.join(xml_path, image_name + '.xml')
# if os.path.exists(xml_save_path):
# continue
# print(img_save_path)
det = dn.detect_ext(net, meta, bytes(image_path, 'utf-8'), thresh)
# det_fr = dn.detect_ext(net_fr, meta_fr, bytes(image_path,'utf-8'),thresh)
img = cv2.imread(image_path)
if img is None:
print('Can not open image')
continue
h, w, c = img.shape
writeXml(xml_save_path, w, h, image_name, det)
dn.free_net(net)
def batch_analysis(weights_list_file, cfg_file, meta_file, image_list_file,
thresh, iou_thresh, result_dir):
image_list = LoadFileList(image_list_file)
image_num = len(image_list)
weights_list = LoadFileList(weights_list_file)
meta = dn.load_meta(meta_file)
# print(meta.classes)
object_type = [
meta.names[i].decode('utf-8').strip() for i in range(meta.classes)
]
result = []
for weights in weights_list:
weights_name = getFileName(weights)
result_path = os.path.join(result_dir, weights_name)
if not os.path.exists(result_path):
os.mkdir(result_path)
net = dn.load_net(cfg_file, bytes(weights, 'utf-8'), 0)
# detect result and save to text
for j, image_path in enumerate(image_list):
print('detect: ' + str(j + 1) + '/' + str(len(image_list)))
label_path = imagePath2labelPath(image_path)
image_name = getFileName(image_path)
det_save_path = os.path.join(result_path, image_name + '.txt')
# print(img_save_path)
det = dn.detect_ext(net, meta, bytes(image_path, 'utf-8'), thresh)
# save detection result to text
saveDetRes(det[0], det_save_path, object_type)
time.sleep(0.001)
# dn.free_net(net)
# campare label and detection result
for i, objtype in enumerate(object_type):
# if objtype != 'fr':
# continue
total_label = 0
total_detect = 0
total_corr = 0
total_iou = 0
cmp_result = []
# print(objtype)
for j, image_path in enumerate(image_list):
label_path = imagePath2labelPath(image_path)
image_name = getFileName(image_path)
img_save_path = os.path.join(result_path, image_name + '.jpg')
det_save_path = os.path.join(result_path, image_name + '.txt')
# print(img_save_path)
label = []
if os.path.exists(label_path):
label = LoadLabel(label_path, object_type)
# save detection result to text
det = readDetRes(det_save_path)
for d in det:
if d[0] > len(object_type) - 1:
d[0] = ' '
continue
d[0] = object_type[d[0]]
# print(label)
# print(det)
# print('')
if i > 0:
image_path = img_save_path
# print(j,image_path)
img = cv2.imread(image_path)
if img is None:
print(
"load image error&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&"
)
# print(img.shape)
cmp_res = cdl.cmp_data(objtype, det, label, thresh, iou_thresh,
img)
cmp_res.update_tracking({'image_name': image_name})
total_corr += cmp_res['correct']
total_iou += cmp_res['avg_iou'] * cmp_res['label_num']
cmp_result.append(cmp_res)
print("%s: %d/%d label: %d detect: %d correct: %d recall: %f avg_iou: %f accuracy: %f precision: %f\n" % \
(str(objtype),j+1,image_num,cmp_res['label_num'],cmp_res['detect_num'],\
cmp_res['correct'],cmp_res['recall'],cmp_res['avg_iou'],\
cmp_res['accuracy'],cmp_res['precision']))
total_label += cmp_res['label_num']
total_detect += cmp_res['detect_num']
cv2.imwrite(img_save_path, img)
time.sleep(0.001)
#数据集分析结果
avg_recall = 0
if total_label > 0:
avg_recall = total_corr / float(total_label)
avg_iou = 0
if total_iou > 0:
avg_iou = total_iou / total_label
avg_acc = 0
if total_label + total_detect - total_corr > 0:
avg_acc = float(total_corr) / (total_label + total_detect -
total_corr)
avg_precision = 0
if total_detect > 0:
avg_precision = float(total_corr) / total_detect
total_result = [
total_label, total_detect, total_corr, avg_recall, avg_iou,
avg_acc, avg_precision
]
cdl.ExportAnaRes(objtype, cmp_result, total_result, image_path,
result_path)
print("total_label: %d total_detect: %d total_corr: %d recall: %f average iou: %f accuracy: %f precision: %f \n" % \
(total_result[0],total_result[1],total_result[2],total_result[3],total_result[4],total_result[5],total_result[6]))
result.append([weights_name] + [objtype] + total_result)
# dn.free_net(net)
cdl.ExportAnaResAll(result, result_dir)
time.sleep(0.001)
def batch_analysis(weights_list_file, image_list_file, thresh, iou_thresh,
result_dir):
image_list = LoadFileList(image_list_file)
image_num = len(image_list)
weights_list = LoadFileList(weights_list_file)
result = []
for weights in weights_list:
weights_name = getFileName(weights)
# print('weights_name: ',weights)
meta_file, cfg_file = getMetaCfgName(weights)
# meta = dn.load_meta(meta_file)
# net = dn.load_net(cfg_file,bytes(weights,'utf-8'),0)
# 选择对应的dn
meta = dn.load_meta(meta_file)
net = dn.load_net(cfg_file, bytes(weights, 'utf-8'), 0)
object_type = [
meta.names[i].decode('utf-8').strip() for i in range(meta.classes)
]
result_path = os.path.join(result_dir, weights_name)
if not os.path.exists(result_path):
os.mkdir(result_path)
# detect result and save to text
timeall = 0
for j, image_path in enumerate(image_list):
print('detect: ' + str(j + 1) + '/' + str(len(image_list)))
label_path = imagePath2labelPath(image_path)
image_name = getFileName(image_path)
det_save_path = os.path.join(result_path, image_name + '.txt')
# det = dn.detect_ext(net, meta, bytes(image_path,'utf-8'),thresh)
# 选择对应的dn
det, time1 = dn.detect_ext(net, meta, bytes(image_path, 'utf-8'),
thresh)
timeall = timeall + time1
# save detection result to text
saveDetRes(det, det_save_path, object_type)
time.sleep(0.001)
print('xxxxxxxxxxx', 'FPS, ', len(image_list) / timeall)
# dn.free_net(net)
# campare label and detection result
for i, objtype in enumerate(object_type):
# if objtype != 'fr':
# continue
total_label = 0
total_detect = 0
total_corr = 0
total_iou = 0
cmp_result = []
det_ = []
annopath = []
detall = [['name', 'obj_type', 'score', 0, 0, 0,
0]] # 此处为xywh(中心)，应该变为xmin,ymin,xmax,ymax
imagesetfile = []
for j, image_path in enumerate(image_list):
label_path = imagePath2labelPath(image_path)
image_name = getFileName(image_path)
imagesetfile.append(image_name)
img_save_path = os.path.join(result_path, image_name + '.jpg')
det_save_path = os.path.join(result_path, image_name + '.txt')
# detpath.append(det_save_path)
annopath.append(label_path)
# print(img_save_path)
label = []
if os.path.exists(label_path):
label = LoadLabel(label_path, object_type)
# save detection result to text
det = readDetRes(det_save_path)
for d in det:
if d[0] > len(object_type) - 1:
d[0] = ' '
continue
d[0] = object_type[d[0]]
for d in det:
xmin = float(copy.deepcopy(
d[2])) - float(copy.deepcopy(d[4])) / 2.0
ymin = float(copy.deepcopy(
d[3])) - float(copy.deepcopy(d[5])) / 2.0
xmax = float(copy.deepcopy(
d[2])) + float(copy.deepcopy(d[4])) / 2.0
ymax = float(copy.deepcopy(
d[3])) + float(copy.deepcopy(d[5])) / 2.0
# 该文件格式：imagename1 type confidence xmin ymin xmax ymax
d_ = [image_name, d[0], d[1], xmin, ymin, xmax, ymax]
det_.append(d_)
if len(det_) != 0:
detall = numpy.vstack((detall, det_))
det_ = []
if i > 0:
image_path = img_save_path
# print(j,image_path)
img = cv2.imread(image_path)
if img is None:
print(
"load image error&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&"
)
continue
cmp_res = cdl.cmp_data(objtype, det, label, thresh, iou_thresh,
img)
cmp_res.update_tracking({'image_name': image_name})
total_corr += cmp_res['correct']
total_iou += cmp_res['avg_iou'] * cmp_res['label_num']
cmp_result.append(cmp_res)
print("%s: %d/%d label: %d detect: %d correct: %d recall: %f avg_iou: %f accuracy: %f precision: %f\n" % \
(str(objtype),j+1,image_num,cmp_res['label_num'],cmp_res['detect_num'],\
cmp_res['correct'],cmp_res['recall'],cmp_res['avg_iou'],\
cmp_res['accuracy'],cmp_res['precision']))
total_label += cmp_res['label_num']
total_detect += cmp_res['detect_num']
cv2.imwrite(img_save_path, img)
img = []
time.sleep(0.001)
# 求出AP值
ap = 0
# detall = numpy.delete(detall, 0, axis = 0)
# det_objtype = [obj for obj in detall if obj[1] == objtype]
# if len(det_objtype) == 0:
# ap = 0
# else:
# ap = voc_eval(det_objtype, annopath, imagesetfile, objtype, iou_thresh)
# detall=[]
#数据集分析结果
avg_recall = 0
if total_label > 0:
avg_recall = total_corr / float(total_label)
avg_iou = 0
if total_iou > 0:
avg_iou = total_iou / total_label
avg_acc = 0
if total_label + total_detect - total_corr > 0:
avg_acc = float(total_corr) / (total_label + total_detect -
total_corr)
avg_precision = 0
if total_detect > 0:
avg_precision = float(total_corr) / total_detect
total_result = [
total_label, total_detect, total_corr, avg_recall, avg_iou,
avg_acc, avg_precision
]
cdl.ExportAnaRes(objtype, cmp_result, total_result, image_path,
result_path)
print("total_label: %d total_detect: %d total_corr: %d recall: %f average iou: %f accuracy: %f precision: %f ap: %f\n" % \
(total_result[0],total_result[1],total_result[2],total_result[3],total_result[4],total_result[5],total_result[6],ap))
result.append([weights_name] + [objtype] + total_result +
[float(ap)])
cdl.ExportAnaResAll(result, result_dir)
time.sleep(0.001)
def batch_analysis(meta_file, cfg_file, wgt_file, thresh, nms, src_path,
dst_path):
image_list = listdir(src_path)
image_list.sort()
image_num = len(image_list)
meta = dn.load_meta(meta_file)
object_type = [
meta.names[i].decode('utf-8').strip() for i in range(meta.classes)
]
net = dn.load_net(cfg_file, wgt_file, 0)
move_count = 0
boxes_last = []
for j, image_path in enumerate(image_list):
print(str(j) + '/' + str(image_num) + " moved: " + str(move_count))
# print(image_path)
try:
img = cv2.imread(image_path)
except:
print(
'can not read image******************************************')
continue
h, w = img.shape[:2]
image_name = getFileName(image_path)
print("image_name", image_name)
image_name = image_name.replace('(', '1_')
image_name = image_name.replace(')', '_1')
img_save_path = os.path.join(dst_path, image_name + '.jpg')
# print(img_save_path)
det = dn.detect_ext(net, meta, bytes(image_path, 'utf-8'), thresh)
boxes = []
is_move_file = False
if j % 20 == 0: #20数值越大 比对iou的间隔越大
is_move_file = True
for d in det:
# try:
# img = cv2.imread(image_path)
# except:
# print('can not read image******************************************')
# continue
# h,w = img.shape[:2]
print("d", d)
boxes.append(d[2:])
bw = d[4] * w
bh = d[5] * h
# if bw < 20 or bh < 20:
# print("bw or bh is less than 20")
# continue
# obj_type = d[0]
# if obj_type == 'tricycle':
# print("tricycle ************************************************")
# is_move_file = True
# break
# elif obj_type == 'car':
# if bw*bh/(w*h) > 0.25:
# print("big car ....................................................")
# is_move_file = True
# break
if boxes_last != [] and boxes != []:
iou = batch_iou(boxes_last, boxes, w, h)
# print('iou: '+str(iou))
if iou > 0.6:
print('batch iou: ' + str(iou))
is_move_file = False
print("iou^^^^^^^^^^^^^^^^^^^^^^^^^")
# continue
if is_move_file:
move_count += 1
if not os.path.exists(img_save_path):
mymovefile(image_path, img_save_path)
boxes_last = boxes
dn.free_net(net)
