官网直接python接口

detect.py

# video detecting is written by my self 
import cv2
from google.colab.patches import cv2_imshow
import PIL
from PIL import Image
import argparse
import os from ctypes import *
import math
import random
import time
import numpy as np
import matplotlib.pyplot as plt# def convertBack(x, y, w, h):
#     xmin = int(round(x - (w / 2)))
#     xmax = int(round(x + (w / 2)))
#     ymin = int(round(y - (h / 2)))
#     ymax = int(round(y + (h / 2)))
#     return xmin, ymin, xmax, ymaxdef array_to_image(arr):# need to return old values to avoid python freeing memoryarr = arr.transpose(2,0,1)c = arr.shape[0]h = arr.shape[1]w = arr.shape[2]arr = (arr/255.0).flatten()data = dn.c_array(dn.c_float, arr)im = dn.IMAGE(w,h,c,data)return im########################################################################### Remember to change it 
import os
os.sys.path.append('/content/drive/My Drive/yolo/darknet/python')
from darknet import *
import darknet as dn# OpenCVif __name__ == "__main__":parser = argparse.ArgumentParser(description='yolo3-tiny? for detection')#parser.add_argument('--image', '-I', help='image 属性，非必要参数,但是有默认值', default=1)parser.add_argument('--out', help='out,输出视频的地址',default ="/content/drive/My Drive/yolo/darknet/video/video_out/out.mp4")parser.add_argument('--path',  help='path,输入需要detect地址', required=True)parser.add_argument('--cfg', help = 'such as yolov3.cfg',required=True)parser.add_argument('--weights', help = 'such as yolov3.weights',required=True)parser.add_argument('--meta', help = 'such as mask.data',required=True)args = parser.parse_args()# 注意一定要在路径前加上b， 由于c语言 remember to add b before string due to c format#net = dn.load_net(b"cfg/yolov3.cfg", b"weights/mask.weights", 0)net = dn.load_net(bytes(args.cfg, encoding='utf-8'), bytes(args.weights, encoding='utf-8'), 0)meta=dn.load_meta(bytes(args.meta, encoding='utf-8')) # the  .data doc records class and train test info#read video  all by myself cap = cv2.VideoCapture(args.path)total_frames = cap.get(cv2.CAP_PROP_FRAME_COUNT)width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))fps = cap.get(cv2.CAP_PROP_FPS)print("Frames per second using video.get(cv2.CAP_PROP_FPS) : {0}".format(fps))# Fourcc decoding typefourcc = cv2.VideoWriter_fourcc(*'mp4v') #or 'XVID'out = cv2.VideoWriter(args.out, fourcc, fps, (width, height))count=0while(1):count=count+1print(count)if count==1*fps:breakret, img = cap.read() ## 返回两个值，ret表示读取是否成功，frame/img为读取的帧内容#img = np.array(img)if ret: #change cv array to IMAGE object#print(type(img))im=dn.ndarray_to_image(img) # image structurer = dn.detect(net, meta, im)for i in range(len(r)):#print(r[i])x1 = r[i][2][0]-r[i][2][2]/2y1 = r[i][2][1]-r[i][2][3]/2x2 = r[i][2][0]+r[i][2][2]/2y2 = r[i][2][1]+r[i][2][3]/2label = r[i][0]score = r[i][1]print(label.decode())pt1 = (int(x1), int(y1))pt2 = (int(x2), int(y2))#print(label)#cv2.rectangle(img, pt1, pt2, (0, 255, 0), 1)cv2.rectangle(img,(int(x1),int(y1)),(int(x2),int(y2)),(0,255,0),2)cv2.putText(img, str(label.decode()), (int(x1),int(y1)),fontFace=cv2.FONT_HERSHEY_SIMPLEX, fontScale =0.8, color=(0,255,255), thickness=2)# plt.imshow(img)# plt.show()out.write(img) # output videoelse: break# k==cv2.waitKey(0)# if  k==27 & 0xFF == ord('q'):#     breakcap.release()out.release()cv2.destroyAllWindows()

yolov3-tiny.cfg

[net] #[net]主要记载了，测试和训练的参数
# Testing  
#batch=1
#subdivisions=1
# Training
batch=64
subdivisions=8 #把batch分组输入减少内存压力，colab上应该不用顾虑。width=416     #默认大小416*416, 3通道 -> (3 416 416)
height=416
channels=3
momentum=0.9
decay=0.0005
angle=180  #-180-180   #图片随机旋转多少(-度 ~ +度)，数据增强
saturation = 1.5    
exposure = 1.5
hue=.1learning_rate=0.05  #0.001
burn_in=1000  #在迭代次数小于burn_in时，其学习率的更新有一种方式，大于burn_in时，才采用policy的更新方式
max_batches = 205000   #500200
policy=steps
#steps=100,1000,1500,2000, 10000, 15000   # 400000-450000 在多少迭代后，进行learning rate 缩放scales因子
steps=1000,3000,7000#, 12000 
scales=.1,.5,.5, #.5#, .1[convolutional]
batch_normalize=1
filters=16
size=3
stride=1
pad=1
activation=leaky[maxpool]
size=2
stride=2[convolutional]
batch_normalize=1
filters=32
size=3
stride=1
pad=1
activation=leaky[maxpool]
size=2
stride=2[convolutional]
batch_normalize=1
filters=64
size=3
stride=1
pad=1
activation=leaky[maxpool]
size=2
stride=2[convolutional]
batch_normalize=1
filters=128
size=3
stride=1
pad=1
activation=leaky[maxpool]
size=2
stride=2[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky[maxpool]
size=2
stride=2[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky[maxpool]
size=2
stride=1[convolutional]
batch_normalize=1
filters=1024
size=3
stride=1
pad=1
activation=leaky###########[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky[convolutional]
size=1
stride=1
pad=1
filters=21
activation=linear[yolo]
mask = 3,4,5
anchors = 10,14,  23,27,  37,58,  81,82,  135,169,  344,319
classes=2
num=6        # 6 box for a grid cell一个格子预测几个回归框，这一层是3,4,5这三个框
jitter=.3     # RGB转成HSV然后，三个值进行5%-10%的随机增减然后变回RGB
ignore_thresh = .6   # 0.7参与计算的IOU最低值
truth_thresh = 1
random=1[route]  #跨连接层
layers = -4[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky[upsample]
stride=2[route]
layers = -1, 8[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky[convolutional]
size=1
stride=1
pad=1
filters=21
activation=linear[yolo]
mask = 0,1,2
anchors = 10,14,  23,27,  37,58,  81,82,  135,169,  344,319
classes=2
num=6
jitter=.3
ignore_thresh = .6
truth_thresh = 1
random=1

colab训练截图

!sudo apt-get update
!sudo apt-get install dos2unix
!dos2unix cfg/mask_det.data
!dos2unix cfg/mask_det.names
!dos2unix cfg/yolov3-tiny.cfg
!dos2unix scripts/f_train.txt

!pwd
%cd /content/drive/My Drive/yolo/darknet!chmod u+x darknet
!./darknet detector train cfg/mask_det.data cfg/yolov3-tiny.cfg /content/mydirver/yolo/darknet/weights/yolov3-tiny.backup

测试之前将训练师的CUDNN关闭，否则会没有框

# when use detect,set cudnn=0 
%cd /content/mydirver/yolo/darknet
!make clean
!make

%matplotlib inline
!python3 det.py --path '/content/drive/My Drive/yolo/darknet/video/1.mp4' --cfg '/content/drive/My Drive/yolo/darknet/cfg/yolov3-tiny.cfg' --weights '/content/drive/My Drive/yolo/darknet/weights/yolov3-tiny_60000.weights' --meta "/content/drive/My Drive/yolo/darknet/cfg/mask_det.data"

!pwd
%cd /content/drive/My Drive/yolo/darknet
!python3 detect_img.py --image '1' --path '/content/mydirver/yolo/darknet/data/mask3.jpg' --cfg '/content/drive/My Drive/yolo/darknet/cfg/yolov3-tiny.cfg' --weights '/content/drive/My Drive/yolo/darknet/weights/yolov3-tiny.backup' --meta "/content/drive/My Drive/yolo/darknet/cfg/mask_det.data"import cv2
from google.colab.patches import cv2_imshow
img = cv2.imread('/content/drive/My Drive/yolo/darknet/prediction.jpg')
print(img.shape)
cv2_imshow(img)

!pwd
%cd /content/drive/My Drive/yolo/darknet
!python3 detect_img.py --image '1' --path '/content/mydirver/yolo/darknet/data/mask2.jpg' --cfg '/content/drive/My Drive/yolo/darknet/cfg/yolov3-tiny.cfg' --weights '/content/drive/My Drive/yolo/darknet/weights/yolov3-tiny.backup' --meta "/content/drive/My Drive/yolo/darknet/cfg/mask_det.data"import cv2
from google.colab.patches import cv2_imshow
img = cv2.imread('/content/drive/My Drive/yolo/darknet/prediction.jpg')
print(img.shape)
cv2_imshow(img)

谢谢各位老爷见证我的菜鸡成长之路！

传送门