def detect(save_img=False):
#save_img=False
# 获取输出文件夹,输入源,权重,参数等参数
out, source, weights, view_img, save_txt, imgsz = \
opt.output, opt.source, opt.weights, opt.view_img, opt.save_txt, opt.img_size
#webcam获取source的信息返回true表示是?
webcam = source.isnumeric() or source.startswith('rtsp') or source.startswith('http') or source.endswith('.txt')
#print(webcam) webcam web或者摄像头
# Initialize
set_logging()
device = select_device(opt.device)
#out默认输出目录为inference / output,存在就删除输出目录,再建目录
if os.path.exists(out):
shutil.rmtree(out) # delete output folder
os.makedirs(out) # make new output folder
half = device.type != 'cpu' # half precision only supported on CUDA
# Load model
print("-----weights:",weights)
model = attempt_load(weights, map_location=device) # load FP32 model,此处加载模型该模型在程序同级目录下
#---装载------------------------------------------------------------
imgsz = check_img_size(imgsz, s=model.stride.max()) # check img_size
if half:
model.half() # to FP16
# Second-stage classifier# 设置第二次分类,默认不使用
classify = False
if classify:
modelc = load_classifier(name='resnet101', n=2) # initialize
modelc.load_state_dict(torch.load('weights/resnet101.pt', map_location=device)['model']) # load weights
modelc.to(device).eval()
# Set Dataloader# 通过不同的输入源来设置不同的数据加载方式
vid_path, vid_writer = None, None
if webcam:
view_img = True
#如果网络的输入数据维度或类型上变化不大,也就是每次训练的图像尺寸都是一样的时候,设置 torch.backends.cudnn.benchmark = true 可以增加运行效率;
cudnn.benchmark = True # set True to speed up constant image size inference可加速图像预测
dataset = LoadStreams(source, img_size=imgsz)
#获取视频信息,线程抓取图片dataset类中imgs[0]是0个摄像头的图片,LoadStreams是迭代类---》dataset是一个迭代器
#dataset获取的数据是(sources, img, img0, None)
# img0是原始数据源获取的图片列表,img是处理后的图片列表,-----图像可能是是与设备同宽, 在上下添加黑边的显示模式;
else:
save_img = True
dataset = LoadImages(source, img_size=imgsz) #?????????????????????????
#print("dataset=",dataset)
# Get names and colors
names = model.module.names if hasattr(model, 'module') else model.names
colors = [[random.randint(0, 255) for _ in range(3)] for _ in range(len(names))]
# Run inference# 进行一次前向推理,测试程序是否正常
t0 = time.time()
img = torch.zeros((1, 3, imgsz, imgsz), device=device) # init img
_ = model(img.half() if half else img) if device.type != 'cpu' else None # run once 测试程序是否正常(非cpu)
#-----------------摄像头从此处开始反复循环-dataset为迭代器类--------------------------------
for path, img, im0s, vid_cap in dataset: #见上面
#path 如果是摄像头就是摄像头编号,为保存识别准备文件名,例如:在inference\output文件夹中0.txt
#把图片转为tensor,放到device上去计算
img = torch.from_numpy(img).to(device)
img = img.half() if half else img.float() # uint8 to fp16/32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
if img.ndimension() == 3:
img = img.unsqueeze(0) #扩大一个维度
# Inference
t1 = time_synchronized() #等待当前设备上所有流中的所有核心完成然后返回时间(自定义)
"""
?????前向传播?????? 返回pred的shape是(1, num_boxes, 5+num_class)
h,w为传入网络图片的长和宽,注意dataset在检测时使用了矩形推理,所以这里h不一定等于w
num_boxes = h/32 * w/32 + h/16 * w/16 + h/8 * w/8
pred[..., 0:4]为预测框坐标
预测框坐标为xywh(中心点+宽长)格式
pred[..., 4]为objectness置信度
pred[..., 5:-1]为分类结果,5:-1 5到倒数第一个
"""
pred_all = model(img, augment=opt.augment)
#print("pred_all[1].len=",len(pred_all[1]))
#pred_all结构为长度为2的tuple
#pred_all[0]=(tensor[1, 20160, 7]) 20160=3*(3个网格点数),3个网格点数=64*80+32*40+16*4)
# pred_all[0]为所有网格点预测结果汇总,pred_all[1]为原始预测结果
#pred_all[1]=(tensor1[shape为1, 3, 64, 80, 7]),tensor2([1, 3, 64/2, 80/2, 7]),tensor3([1, 3, 64/4, 80/4, 7])
pred = pred_all[0] #取出3个网格预测结果,torch.Size([1, 20160, 7])
# print("-------------",pred.shape)
# input("pred")
# Apply NMS
"""
pred:前向传播的输出
conf_thres:置信度阈值
iou_thres:iou阈值
classes:是否只保留特定的类别
agnostic:进行nms是否也去除不同类别之间的框
经过nms之后,预测框格式:xywh-->xyxy(左上角右下角)
pred是一个列表list[torch.tensor],长度为batch_size
每一个torch.tensor的shape为(num_boxes, 6),6的内容为box+置信度+类别
"""
# 下面的pred.shape=[1,tensor(x,6),...] x是预测框个数,6的内容为box位置(4个)+置信度+类别(该函数取出预测结果)
pred = non_max_suppression(pred, opt.conf_thres, opt.iou_thres, classes=opt.classes, agnostic=opt.agnostic_nms)
# print("------------pred\n",pred,"\n")
# input("----------------")
#pred经过非极大抑制后去掉了多余预测结果结构变为 [tensor[[x1,y1,x2,y2,概率,类别],[...],...]]
t2 = time_synchronized()
# Apply Classifier
if classify:
pred = apply_classifier(pred, modelc, img, im0s)
#mm=list(enumerate(pred))
#print("-----------mm\n",mm)
# Process detections
for i, det in enumerate(pred): # detections per image, pred 是所有预测结果,[[x0,y0,x1,y1,置信度,类别],....]
#det是每张图的多个预测结果,pred是多张图的预测结果
#此循环,如果是视频,只循环一次,pred就是一张图预测结果
# print(len(pred))
# input("pred len---------------")
if webcam: # batch_size >= 1#是摄像头或者视频
#如果是视频 path=["0"]或者["1"],im0s=[一张原始图片]
# print("---path[i]:",i,path[i])
p, s, im0 = path[i], '%g: ' % i, im0s[i].copy()
else:#如果是图片path=[多张图片路径],im0s多张图片数据
p, s, im0 = path, '', im0s
save_path = str(Path(out) / Path(p).name)#保存输出结果的路径
#print("save:=",save_path)
# print("dataset.mode=____",dataset.mode)
txt_path = str(Path(out) / Path(p).stem) + ('_%g' % dataset.frame if dataset.mode == 'video' else '')
s += '%gx%g ' % img.shape[2:] # print string
gn = torch.tensor(im0.shape)[[1, 0, 1, 0]] # normalization gain whwh ,gn=[w,h,w,h]图片的宽高
if det is not None and len(det):
# Rescale boxes from img_size to im0 size
det[:, :4] = scale_coords(img.shape[2:], det[:, :4], im0.shape).round()
# print(det)
# input("det")
# Print results,,, c取出来的就是类别,而且唯一,下面输出各类别的预测数
for c in det[:, -1].unique(): #det.shape=(n,6),第二维度最后一个数(是类别),unique()去除det倒数第一行的重复元素,且排序
n = (det[:, -1] == c).sum() # detections per class每类的检测数量
s += '%g %ss, ' % (n, names[int(c)]) # add to string
# Write results#此处循环画框--------------------------------------
for *xyxy, conf, cls in reversed(det): #?reversed(det)啥意思
# print(reversed(det))
# input("---------reversed(det-------")
if save_txt: # Write to file
xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) / gn).view(-1).tolist() # normalized xywh
with open(txt_path + '.txt', 'a') as f:
f.write(('%g ' * 5 + '\n') % (cls, *xywh)) # label format
if save_img or view_img: # Add bbox to image
label = '%s %.2f' % (names[int(cls)], conf)
#图片上标记框框
#img0是视频获取的图片
plot_one_box(xyxy, im0, label=label, color=colors[int(cls)], line_thickness=3)
#---------------------------------------------
myxyxy=(xyxy[0].item(), xyxy[1].item(), xyxy[2].item(), xyxy[3].item())
myp=conf.item()
myclass=cls.item()
print("类别、置信度、坐标::",myclass,myp,myxyxy)
#-------------------------------输出坐标------------------------------------------------------------
# Print time (inference + NMS)
print('%sDone. (%.3fs)' % (s, t2 - t1))
# Stream results
if view_img:
#此处显示视频流和框的结果
cv2.imshow(p, im0)
if cv2.waitKey(1) == ord('q'): # q to quit
raise StopIteration
# Save results (image with detections)
if save_img:#如果保存结果
if dataset.mode == 'images': #如果是图
cv2.imwrite(save_path, im0)
else:#如果不是图,是视频
if vid_path != save_path: # new video
vid_path = save_path
if isinstance(vid_writer, cv2.VideoWriter):
vid_writer.release() # release previous video writer
fourcc = 'mp4v' # output video codec
fps = vid_cap.get(cv2.CAP_PROP_FPS)
w = int(vid_cap.get(cv2.CAP_PROP_FRAME_WIDTH))
h = int(vid_cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
vid_writer = cv2.VideoWriter(save_path, cv2.VideoWriter_fourcc(*fourcc), fps, (w, h))
vid_writer.write(im0)
if save_txt or save_img:
print('Results saved to %s' % Path(out))
if platform.system() == 'Darwin' and not opt.update: # MacOS
os.system('open ' + save_path)
print('Done. (%.3fs)' % (time.time() - t0))