目標
使用 darknet (https://github.com/pjreddie/darknet) 自帶的 python 接口處理圖片和視頻。
project 下載
git clone https://github.com/pjreddie/darknet
cd darknet
#改一些配置 ,具體操作見 我的上一份博客的結尾部分
#（https://blog.csdn.net/qq_20241587/article/details/111176541）
make處理單張圖片
即：指定一張圖片的路徑，指定檢測結果新圖片的存放位置，進行 model檢測+畫框+另存爲新圖片.
下圖中，左邊是處理前的樣子，右邊是處理後的樣子，兩張圖位置在代碼中指定
代碼如下。
核心改動處 ：
1 # lib = CDLL("libdarknet.so", RTLD_GLOBAL) , 改成自己的項目的具體地址
2 使用 cv2.rectangle 畫框框， 使用cv2.putText 放文字，爲了避免框框和文字交叉，我加了一丟丟的偏移量。
其中 yolov3的輸出是 ：
label_i = box_i[0] #標籤
prob_i = box_i[1] #標籤置信度
x_ = box_i[2][0]
y_ = box_i[2][1]
w_ = box_i[2][2]
h_ = box_i[2][3] # bbox信息(x,y,w,h)爲物體的中心位置相對格子位置的偏移及寬度和高度,
cv2.rectangle(image, (int(x_ - w_ / 2), int(y_ - h_ / 2)),
(int(x_ + w_ / 2), int(y_ + h_ / 2)),
color, line_type)
cv2.putText(image, text_, (int(x_ - w_ / 2 - 5), int(y_ - h_ / 2 - 5)), cv2.FONT_HERSHEY_DUPLEX, 0.7, color,
2)
from ctypes import *
import math
import random
import cv2
import os
def sample(probs):
s = sum(probs)
probs = [a / s for a in probs]
r = random.uniform(0, 1)
for i in range(len(probs)):
r = r - probs[i]
if r <= 0:
return i
return len(probs) - 1
def c_array(ctype, values):
arr = (ctype * len(values))()
arr[:] = values
return arr
class BOX(Structure):
_fields_ = [("x", c_float),
("y", c_float),
("w", c_float),
("h", c_float)]
class DETECTION(Structure):
_fields_ = [("bbox", BOX),
("classes", c_int),
("prob", POINTER(c_float)),
("mask", POINTER(c_float)),
("objectness", c_float),
("sort_class", c_int)]
class IMAGE(Structure):
_fields_ = [("w", c_int),
("h", c_int),
("c", c_int),
("data", POINTER(c_float))]
class METADATA(Structure):
_fields_ = [("classes", c_int),
("names", POINTER(c_char_p))]
lib = CDLL("/home/jiantang/桌面/enn/workcode/yoloV3/github/darknet/libdarknet.so", RTLD_GLOBAL)
# lib = CDLL("libdarknet.so", RTLD_GLOBAL)
lib.network_width.argtypes = [c_void_p]
lib.network_width.restype = c_int
lib.network_height.argtypes = [c_void_p]
lib.network_height.restype = c_int
predict = lib.network_predict
predict.argtypes = [c_void_p, POINTER(c_float)]
predict.restype = POINTER(c_float)
set_gpu = lib.cuda_set_device
set_gpu.argtypes = [c_int]
make_image = lib.make_image
make_image.argtypes = [c_int, c_int, c_int]
make_image.restype = IMAGE
get_network_boxes = lib.get_network_boxes
get_network_boxes.argtypes = [c_void_p, c_int, c_int, c_float, c_float, POINTER(c_int), c_int, POINTER(c_int)]
get_network_boxes.restype = POINTER(DETECTION)
make_network_boxes = lib.make_network_boxes
make_network_boxes.argtypes = [c_void_p]
make_network_boxes.restype = POINTER(DETECTION)
free_detections = lib.free_detections
free_detections.argtypes = [POINTER(DETECTION), c_int]
free_ptrs = lib.free_ptrs
free_ptrs.argtypes = [POINTER(c_void_p), c_int]
network_predict = lib.network_predict
network_predict.argtypes = [c_void_p, POINTER(c_float)]
reset_rnn = lib.reset_rnn
reset_rnn.argtypes = [c_void_p]
load_net = lib.load_network
load_net.argtypes = [c_char_p, c_char_p, c_int]
load_net.restype = c_void_p
do_nms_obj = lib.do_nms_obj
do_nms_obj.argtypes = [POINTER(DETECTION), c_int, c_int, c_float]
do_nms_sort = lib.do_nms_sort
do_nms_sort.argtypes = [POINTER(DETECTION), c_int, c_int, c_float]
free_image = lib.free_image
free_image.argtypes = [IMAGE]
letterbox_image = lib.letterbox_image
letterbox_image.argtypes = [IMAGE, c_int, c_int]
letterbox_image.restype = IMAGE
load_meta = lib.get_metadata
lib.get_metadata.argtypes = [c_char_p]
lib.get_metadata.restype = METADATA
load_image = lib.load_image_color
load_image.argtypes = [c_char_p, c_int, c_int]
load_image.restype = IMAGE
rgbgr_image = lib.rgbgr_image
rgbgr_image.argtypes = [IMAGE]
predict_image = lib.network_predict_image
predict_image.argtypes = [c_void_p, IMAGE]
predict_image.restype = POINTER(c_float)
# net_d = load_net(b"../cfg/yolov3.cfg", b"../yolov3.weights", 0)
# meta_d = load_meta(b"../cfg/coco.data")
def classify(net, meta, im):
out = predict_image(net, im)
res = []
for i in range(meta.classes):
res.append((meta.names[i], out[i]))
res = sorted(res, key=lambda x: -x[1])
return res
def detect(net, meta, image, thresh=.5, hier_thresh=.5, nms=.45):
im = load_image(image, 0, 0)
num = c_int(0)
pnum = pointer(num)
predict_image(net, im)
dets = get_network_boxes(net, im.w, im.h, thresh, hier_thresh, None, 0, pnum)
num = pnum[0]
if (nms): do_nms_obj(dets, num, meta.classes, nms);
res = []
for j in range(num):
for i in range(meta.classes):
if dets[j].prob[i] > 0:
b = dets[j].bbox
res.append((meta.names[i], dets[j].prob[i], (b.x, b.y, b.w, b.h)))
res = sorted(res, key=lambda x: -x[1])
free_image(im)
free_detections(dets, num)
return res
def detect_and_boxing(net, meta, b_path, raw_path, save_path,
color=(0.255, 255), line_type=1):
image = cv2.imread(raw_path)
r = detect(net, meta, b_path)
if not len(r) > 0:
print("nothing detected in this picture!")
else:
for i in range(len(r)):
box_i = r[i]
label_i = box_i[0]
prob_i = box_i[1]
x_ = box_i[2][0]
y_ = box_i[2][1]
w_ = box_i[2][2]
h_ = box_i[2][3]
text_ = str(label_i) + "," + str(round(prob_i, 3))
cv2.rectangle(image, (int(x_ - w_ / 2), int(y_ - h_ / 2)),
(int(x_ + w_ / 2), int(y_ + h_ / 2)),
color, line_type)
cv2.putText(image, text_, (int(x_ - w_ / 2 - 5), int(y_ - h_ / 2 - 5)), cv2.FONT_HERSHEY_DUPLEX, 0.7, color,
2)
cv2.imwrite(save_path, image)
print("boxing ", i, " found ", label_i, "with prob = ", prob_i, ", finished!")
print("box position is :", box_i[2])
if __name__ == "__main__":
# net = load_net("cfg/densenet201.cfg", "/home/pjreddie/trained/densenet201.weights", 0)
# im = load_image("data/wolf.jpg", 0, 0)
# meta = load_meta("cfg/imagenet1k.data")
# r = classify(net, meta, im)
# print(r)
net = load_net(b"../cfg/yolov3.cfg", b"../yolov3.weights", 0)
meta = load_meta(b"../cfg/coco.data")
b_path = b"../data/hat_sougou2.jpg"
raw_path = "../data/hat_sougou2.jpg"
save_path = "/home/jiantang/z_test/hat_sougou2.jpg"
detect_and_boxing(net, meta, b_path=b_path, raw_path=raw_path, save_path=save_path)
處理視頻
即：指定一視頻的路徑，指定檢測結果新新品和中間產生的臨時幀的存放位置，進行 幀獲取+ model檢測+畫框+另存爲新幀 + 拼成新video.
（這個視頻43秒，63M，.avi 格式。
產生的框好的新視頻爲 41秒， 55幀每秒， 6.9G，.avi 格式，每幀400kb左右 ）
效果如下，原視頻對比我就不放了，，因爲每秒55幀，所以流暢感沒啥問題。
代碼有一些潛在的問題：
1 .avi格式產生的新視頻size 好大，63M 成了6.9G ，而且這個代碼僅支持.avi格式。
2 因爲是一個視頻，不知道原視頻幀率，所以新視頻指定幀率後，時長有一丟丟差異。
3 處理速度感人 （中間有很多幀s的磁盤讀寫操作，，嚴重拉垮了速度。 model detect 速度使用GPU還是很快的，真要用實時的，不用把幀和新幀存起來，直接走內存display）
代碼如下 （新建一個darknet_video.py文件）：
from ctypes import *
import math
import random
import cv2
import os
def sample(probs):
s = sum(probs)
probs = [a / s for a in probs]
r = random.uniform(0, 1)
for i in range(len(probs)):
r = r - probs[i]
if r <= 0:
return i
return len(probs) - 1
def c_array(ctype, values):
arr = (ctype * len(values))()
arr[:] = values
return arr
class BOX(Structure):
_fields_ = [("x", c_float),
("y", c_float),
("w", c_float),
("h", c_float)]
class DETECTION(Structure):
_fields_ = [("bbox", BOX),
("classes", c_int),
("prob", POINTER(c_float)),
("mask", POINTER(c_float)),
("objectness", c_float),
("sort_class", c_int)]
class IMAGE(Structure):
_fields_ = [("w", c_int),
("h", c_int),
("c", c_int),
("data", POINTER(c_float))]
class METADATA(Structure):
_fields_ = [("classes", c_int),
("names", POINTER(c_char_p))]
lib = CDLL("/home/jiantang/桌面/enn/workcode/yoloV3/github/darknet/libdarknet.so", RTLD_GLOBAL)
# lib = CDLL("libdarknet.so", RTLD_GLOBAL)
lib.network_width.argtypes = [c_void_p]
lib.network_width.restype = c_int
lib.network_height.argtypes = [c_void_p]
lib.network_height.restype = c_int
predict = lib.network_predict
predict.argtypes = [c_void_p, POINTER(c_float)]
predict.restype = POINTER(c_float)
set_gpu = lib.cuda_set_device
set_gpu.argtypes = [c_int]
make_image = lib.make_image
make_image.argtypes = [c_int, c_int, c_int]
make_image.restype = IMAGE
get_network_boxes = lib.get_network_boxes
get_network_boxes.argtypes = [c_void_p, c_int, c_int, c_float, c_float, POINTER(c_int), c_int, POINTER(c_int)]
get_network_boxes.restype = POINTER(DETECTION)
make_network_boxes = lib.make_network_boxes
make_network_boxes.argtypes = [c_void_p]
make_network_boxes.restype = POINTER(DETECTION)
free_detections = lib.free_detections
free_detections.argtypes = [POINTER(DETECTION), c_int]
free_ptrs = lib.free_ptrs
free_ptrs.argtypes = [POINTER(c_void_p), c_int]
network_predict = lib.network_predict
network_predict.argtypes = [c_void_p, POINTER(c_float)]
reset_rnn = lib.reset_rnn
reset_rnn.argtypes = [c_void_p]
load_net = lib.load_network
load_net.argtypes = [c_char_p, c_char_p, c_int]
load_net.restype = c_void_p
do_nms_obj = lib.do_nms_obj
do_nms_obj.argtypes = [POINTER(DETECTION), c_int, c_int, c_float]
do_nms_sort = lib.do_nms_sort
do_nms_sort.argtypes = [POINTER(DETECTION), c_int, c_int, c_float]
free_image = lib.free_image
free_image.argtypes = [IMAGE]
letterbox_image = lib.letterbox_image
letterbox_image.argtypes = [IMAGE, c_int, c_int]
letterbox_image.restype = IMAGE
load_meta = lib.get_metadata
lib.get_metadata.argtypes = [c_char_p]
lib.get_metadata.restype = METADATA
load_image = lib.load_image_color
load_image.argtypes = [c_char_p, c_int, c_int]
load_image.restype = IMAGE
rgbgr_image = lib.rgbgr_image
rgbgr_image.argtypes = [IMAGE]
predict_image = lib.network_predict_image
predict_image.argtypes = [c_void_p, IMAGE]
predict_image.restype = POINTER(c_float)
net_d = load_net(b"../cfg/yolov3.cfg", b"../yolov3.weights", 0)
meta_d = load_meta(b"../cfg/coco.data")
def classify(net, meta, im):
out = predict_image(net, im)
res = []
for i in range(meta.classes):
res.append((meta.names[i], out[i]))
res = sorted(res, key=lambda x: -x[1])
return res
def detect(net, meta, image, thresh=.5, hier_thresh=.5, nms=.45):
im = load_image(image, 0, 0)
num = c_int(0)
pnum = pointer(num)
predict_image(net, im)
dets = get_network_boxes(net, im.w, im.h, thresh, hier_thresh, None, 0, pnum)
num = pnum[0]
if (nms): do_nms_obj(dets, num, meta.classes, nms);
res = []
for j in range(num):
for i in range(meta.classes):
if dets[j].prob[i] > 0:
b = dets[j].bbox
res.append((meta.names[i], dets[j].prob[i], (b.x, b.y, b.w, b.h)))
res = sorted(res, key=lambda x: -x[1])
free_image(im)
free_detections(dets, num)
return res
# calc all box , red label for the biggest one ,yellow label for the rest, save the img to a specific path
def detect_and_boxing_default(b_path, raw_path, save_path,
color=(0.255, 255), line_type=1):
print("checking pic...", raw_path)
image = cv2.imread(raw_path)
r = detect(net_d, meta_d, b_path)
if not len(r) > 0:
print("nothing detected in this picture!")
else:
print(len(r), " stuff detected in this picture! boxing...")
print("going to save as :", save_path)
for i in range(len(r)):
box_i = r[i]
label_i = box_i[0]
prob_i = box_i[1]
x_ = box_i[2][0]
y_ = box_i[2][1]
w_ = box_i[2][2]
h_ = box_i[2][3]
text_ = str(label_i) + "," + str(round(prob_i, 3))
cv2.rectangle(image, (int(x_ - w_ / 2), int(y_ - h_ / 2)),
(int(x_ + w_ / 2), int(y_ + h_ / 2)),
color, line_type)
cv2.putText(image, text_, (int(x_ - w_ / 2 - 5), int(y_ - h_ / 2 - 5)), cv2.FONT_HERSHEY_DUPLEX, 0.7, color,
2)
cv2.imwrite(save_path, image)
def video_to_pics(video_path='/home/jiantang/work_data/sample_video.avi',
video_out_path='/home/jiantang/work_data/'):
print("video_to_pics start...")
vc = cv2.VideoCapture(video_path)
c = 1
if vc.isOpened():
rval, frame = vc.read()
else:
print('open error!')
rval = False
count_c = 1
while rval:
rval, frame = vc.read()
if rval:
print("dealing with frame : ", count_c)
cv2.imwrite(video_out_path + str(int(c)) + '.jpg', frame)
c += 1
cv2.waitKey(1)
count_c += 1
vc.release()
print("video_to_pics finished...")
def pics_boxing(pics_path, save_path):
raw_save_path = save_path
print("pics_boxing start...")
print("checking path : ", pics_path)
pics_names = os.listdir(pics_path)
print("found pics num :", len(pics_names))
count_c = 1
for name in pics_names:
print("dealing with pics ", count_c)
raw_path = pics_path + "/" + name
b_path = bytes(raw_path, encoding="utf8")
save_path = raw_save_path + "/" + name
detect_and_boxing_default(b_path, raw_path, save_path)
count_c += 1
print("pics_boxing finished...")
def pics_to_video(pics_path, video_new_path='/home/jiantang/work_data/sample_video_new.avi', ):
print("pics_to_video start...")
print("checking files in :", pics_path)
file_list = os.listdir(pics_path)
# remove non-jpg files, remove .jpg sign
tmp_jpg = []
for name in file_list:
if not name.endswith('.jpg'):
print("found sth called:", name, ", skip it.")
file_list.remove(name)
continue
tmp_jpg.append(name.replace(".jpg", ""))
# sort names
tmp_jpg.sort(key=int)
fourcc = cv2.VideoWriter_fourcc('I', '4', '2', '0') # 設置輸出視頻爲avi格式
# cap_fps是幀率，可以根據隨意設置；size要和圖片的size一樣，但是通過img.shape得到圖像的參數是
# （height，width，channel），但是此處的size要傳的是（width，height），這裏一定要注意注意，
# 不然結果會打不開，提示“無法解碼多工傳送的流”等.比如通過img.shape得到常用的圖片尺寸
# （1080,1920,3），則size設爲（1920,1080）
cap_fps = 50
size = (1920, 1080)
# 設置視頻輸出的參數
video = cv2.VideoWriter(video_new_path, fourcc, cap_fps, size)
# video.write默認保存彩色圖，如果是彩色圖，則直接保存
for name in tmp_jpg:
img_E = cv2.imread(pics_path + "/" + name + ".jpg")
print("reading....")
video.write(img_E)
video.release()
print("pics_to_video finished...")
video_path = '/home/jiantang/work_data/sample_video.avi'
video_out_path = '/home/jiantang/work_data/pics/'
video_out_dir = '/home/jiantang/work_data/pics'
video_out_new_path = '/home/jiantang/work_data/pics_new'
video_new_path = '/home/jiantang/work_data/sample_video_new.avi'
video_to_pics(video_path, video_out_path)
pics_boxing(video_out_dir, video_out_new_path)
pics_to_video(video_out_new_path, video_new_path)