detect.py

import argparse
import os
import platform
import shutil
import time
from pathlib import Path

import cv2
import torch
import torch.backends.cudnn as cudnn
from numpy import random

from models.experimental import attempt_load
from utils.datasets import LoadStreams, LoadImages,LoadImagesEn
from utils.general import (
    check_img_size, non_max_suppression, apply_classifier, scale_labels,
    xyxy2xywh, plot_one_rotated_box, strip_optimizer, set_logging, rotate_non_max_suppression)
from utils.torch_utils import select_device, load_classifier, time_synchronized
from utils.evaluation_utils import rbox2txt


def detect(save_img=False,kld_flag=False):
    '''
    input: save_img_flag
    output(result):
    '''
    # 获取输出文件夹，输入路径，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 = source.isnumeric() or source.startswith(('rtsp://', 'rtmp://', 'http://')) or source.endswith('.txt')

    # Initialize
    set_logging()
    # 获取设备
    device = select_device(opt.device)
    # 移除之前的输出文件夹,并新建输出文件夹
    if os.path.exists(out):
        shutil.rmtree(out)  # delete output folder
    os.makedirs(out)  # make new output folder
    # 如果设备为gpu，使用Float16
    half = device.type != 'cpu'  # half precision only supported on CUDA

    # Load model
    # 加载Float32模型，确保用户设定的输入图片分辨率能整除最大步长s=32(如不能则调整为能整除并返回)
    '''
    model = Model(
                  (model): Sequential(
                                       (0): Focus(...)
                                       (1): Conv(...)
                                            ...
                                       (24): Detect(...)
                    )
    '''
    model = attempt_load(weights, map_location=device)  # load FP32 model
    imgsz = check_img_size(imgsz, s=model.stride.max())  # check img_size

    # 设置Float16
    if half:
        model.half()  # to FP16

    # Second-stage classifier
    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,ManyPi = None, None,False
    if webcam:
        view_img = True
        cudnn.benchmark = True  # set True to speed up constant image size inference
        dataset = LoadStreams(source, img_size=imgsz)
    else:
        save_img = True
        if ManyPi:#多batch_size
            dataset = LoadImagesEn(source, img_size=imgsz)
            data_loader = torch.utils.data.DataLoader(dataset,
                                                     batch_size=16,
                                                     shuffle=False,
                                                     num_workers=8)
        else:
            dataset = LoadImages(source, img_size=imgsz)

    # Get names and colors
    # 获取类别名字    names = ['person', 'bicycle', 'car',...,'toothbrush']
    names = model.module.names if hasattr(model, 'module') else model.names
    # 设置画框的颜色    colors = [[178, 63, 143], [25, 184, 176], [238, 152, 129],....,[235, 137, 120]]随机设置RGB颜色
    colors = [[random.randint(0, 255) for _ in range(3)] for _ in range(len(names))]

    # Run inference
    t0 = time.time()
    # 进行一次前向推理,测试程序是否正常  向量维度（1，3，imgsz，imgsz）
    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

    """
        path 图片/视频路径  'E:\...\bus.jpg'
        img 进行resize+pad之后的图片   1*3*re_size1*resize2的张量 (3,img_height,img_weight)
        img0 原size图片   (img_height,img_weight,3)          
        cap 当读取图片时为None，读取视频时为视频源   
    """
    if ManyPi:
        vid_cap = None
        for ii, (path, img, im0s) in enumerate(data_loader):
            print(img.shape[1:])
            img = img.to(device)  # torch.from_numpy(img).to(device)
            # 图片也设置为Float16
            img = img.half() if half else img.float()  # uint8 to fp16/32
            img /= 255.0  # 0 - 255 to 0.0 - 1.0
            # 没有batch_size的话则在最前面添加一个轴
            if img.ndimension() == 3:
                # (in_channels,size1,size2) to (1,in_channels,img_height,img_weight)
                img = img.unsqueeze(0)  # 在[0]维增加一个维度

            # Inference
            t1 = time_synchronized()
            """
            model:
            input: in_tensor (batch_size, 3, img_height, img_weight)
            output: 推理时返回 [z,x]
            z tensor: [small+medium+large_inference]  size=(batch_size, 3 * (small_size1*small_size2 + medium_size1*medium_size2 + large_size1*large_size2), nc)
            x list: [small_forward, medium_forward, large_forward]  eg:small_forward.size=( batch_size, 3种scale框, size1, size2, [xywh,score,num_classes]) 
            '''

            前向传播 返回pred[0]的shape是(1, num_boxes, nc)
            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 = 3 * h/32 * w/32 + 3 * h/16 * w/16 + 3 * h/8 * w/8
            pred[0][..., 0:4] 预测框坐标为xywh(中心点+宽长)格式
            pred[0][..., 4]为objectness置信度
            pred[0][..., 5:5+nc]为分类结果
            pred[0][..., 5+nc:]为Θ分类结果
            """
            # pred : (batch_size, num_boxes, no)  batch_size=1
            pred = model(img, augment=opt.augment)[0]

            # Apply NMS
            # 进行NMS
            # pred : list[tensor(batch_size, num_conf_nms, [xylsθ,conf,classid])] θ∈[0,179]
            # pred = non_max_suppression(pred, opt.conf_thres, opt.iou_thres, classes=opt.classes, agnostic=opt.agnostic_nms)
            pred = rotate_non_max_suppression(pred, opt.conf_thres, opt.iou_thres, classes=opt.classes,
                                              agnostic=opt.agnostic_nms, without_iouthres=False)
            t2 = time_synchronized()

            # Apply Classifier
            if classify:
                pred = apply_classifier(pred, modelc, img, im0s)

            # Process detections
            for i, det in enumerate(pred):  # i:image index  det:(num_nms_boxes, [xylsθ,conf,classid]) θ∈[0,179]
                if webcam:  # batch_size >= 1
                    p, s, im0 = path[i], '%g: ' % i, im0s[i].copy()
                else:
                    # p, s, im0 = path, '', im0s
                    # img = img.numpy()
                    p, s, im0 = path[i], '%g: ' % i, im0s[i]  # .copy()

                save_path = str(Path(out) / Path(p).name)  # 图片保存路径+图片名字
                txt_path = str(Path(out) / Path(p).stem) + ('_%g' % dataset.frame if dataset.mode == 'video' else '')
                # print(txt_path)
                s += '%gx%g ' % img.shape[2:]  # print string
                gn = torch.tensor(im0.shape)[[1, 0, 1, 0]]  # normalization gain whwh

                if det is not None and len(det):
                    # Rescale boxes from img_size to im0 size
                    det[:, :5] = scale_labels(img.shape[2:], det[:, :5], im0.shape).round()

                    # Print results    det:(num_nms_boxes, [xylsθ,conf,classid]) θ∈[0,179]
                    for c in det[:, -1].unique():  # unique函数去除其中重复的元素，并按元素（类别）由大到小返回一个新的无元素重复的元组或者列表
                        n = (det[:, -1] == c).sum()  # detections per class  每个类别检测出来的素含量
                        s += '%g %ss, ' % (n, names[int(c)])  # add to string 输出‘数量 类别,’

                    # Write results  det:(num_nms_boxes, [xywhθ,conf,classid]) θ∈[0,179]
                    for *rbox, conf, cls in reversed(det):  # 翻转list的排列结果,改为类别由小到大的排列
                        # rbox=[tensor(x),tensor(y),tensor(w),tensor(h),tsneor(θ)] θ∈[0,179]
                        # 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)
                            classname = '%s' % names[int(cls)]
                            conf_str = '%.3f' % conf
                            rbox2txt(rbox, classname, conf_str, Path(p).stem,
                                     str(out + '/result_txt/result_before_merge'))
                            # plot_one_box(rbox, im0, label=label, color=colors[int(cls)], line_thickness=2)
                            plot_one_rotated_box(rbox, im0.cpu().numpy(), label=label, color=colors[int(cls)],
                                                 line_thickness=1,
                                                 pi_format=False)
    else:
        for path, img, im0s, vid_cap in dataset:
            print(img.shape)
            img = torch.from_numpy(img).to(device)
            # 图片也设置为Float16
            img = img.half() if half else img.float()  # uint8 to fp16/32
            img /= 255.0  # 0 - 255 to 0.0 - 1.0
            # 没有batch_size的话则在最前面添加一个轴
            if img.ndimension() == 3:
                # (in_channels,size1,size2) to (1,in_channels,img_height,img_weight)
                img = img.unsqueeze(0)  # 在[0]维增加一个维度

            # Inference
            t1 = time_synchronized()
            """
            model:
            input: in_tensor (batch_size, 3, img_height, img_weight)
            output: 推理时返回 [z,x]
            z tensor: [small+medium+large_inference]  size=(batch_size, 3 * (small_size1*small_size2 + medium_size1*medium_size2 + large_size1*large_size2), nc)
            x list: [small_forward, medium_forward, large_forward]  eg:small_forward.size=( batch_size, 3种scale框, size1, size2, [xywh,score,num_classes]) 
            '''
                   
            前向传播 返回pred[0]的shape是(1, num_boxes, nc)
            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 = 3 * h/32 * w/32 + 3 * h/16 * w/16 + 3 * h/8 * w/8
            pred[0][..., 0:4] 预测框坐标为xywh(中心点+宽长)格式
            pred[0][..., 4]为objectness置信度
            pred[0][..., 5:5+nc]为分类结果
            pred[0][..., 5+nc:]为Θ分类结果
            """
            # pred : (batch_size, num_boxes, no)  batch_size=1
            pred = model(img, augment=opt.augment)[0]

            # Apply NMS
            # 进行NMS
            # pred : list[tensor(batch_size, num_conf_nms, [xylsθ,conf,classid])] θ∈[0,179]
            #pred = non_max_suppression(pred, opt.conf_thres, opt.iou_thres, classes=opt.classes, agnostic=opt.agnostic_nms)
            pred = rotate_non_max_suppression(pred, opt.conf_thres, opt.iou_thres, classes=opt.classes, agnostic=opt.agnostic_nms, without_iouthres=False,kld_flag=kld_flag)
            t2 = time_synchronized()

            # Apply Classifier
            if classify:
                pred = apply_classifier(pred, modelc, img, im0s)

            # Process detections
            for i, det in enumerate(pred):  # i:image index  det:(num_nms_boxes, [xylsθ,conf,classid]) θ∈[0,179]
                if webcam:  # batch_size >= 1
                    p, s, im0 = path[i], '%g: ' % i, im0s[i].copy()
                else:
                    p, s, im0 = path, '', im0s

                save_path = str(Path(out) / Path(p).name)  # 图片保存路径+图片名字
                txt_path = str(Path(out) / Path(p).stem) + ('_%g' % dataset.frame if dataset.mode == 'video' else '')
                #print(txt_path)
                s += '%gx%g ' % img.shape[2:]  # print string
                gn = torch.tensor(im0.shape)[[1, 0, 1, 0]]  # normalization gain whwh

                if det is not None and len(det):
                    # Rescale boxes from img_size to im0 size
                    det[:, :5] = scale_labels(img.shape[2:], det[:, :5], im0.shape).round()

                    # Print results    det:(num_nms_boxes, [xylsθ,conf,classid]) θ∈[0,179]
                    for c in det[:, -1].unique():  # unique函数去除其中重复的元素，并按元素（类别）由大到小返回一个新的无元素重复的元组或者列表
                        n = (det[:, -1] == c).sum()  # detections per class  每个类别检测出来的素含量
                        s += '%g %ss, ' % (n, names[int(c)])  # add to string 输出‘数量 类别,’

                    # Write results  det:(num_nms_boxes, [xywhθ,conf,classid]) θ∈[0,179]
                    for *rbox, conf, cls in reversed(det):  # 翻转list的排列结果,改为类别由小到大的排列
                        # rbox=[tensor(x),tensor(y),tensor(w),tensor(h),tsneor(θ)] θ∈[0,179]
                        # 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)
                            classname = '%s' % names[int(cls)]
                            conf_str = '%.3f' % conf
                            rbox2txt(rbox, classname, conf_str, Path(p).stem, str(out + '/result_txt/result_before_merge'))
                            #plot_one_box(rbox, im0, label=label, color=colors[int(cls)], line_thickness=2)
                            #不保存图片
                            plot_one_rotated_box(rbox, im0, label=label, color=colors[int(cls)], line_thickness=1,
                                                 pi_format=False)

                # Print time (inference + NMS)
                print('%sDone. (%.3fs)' % (s, t2 - t1))

                # Stream results 播放结果
                if view_img:
                    cv2.imshow(p, im0)
                    if cv2.waitKey(1) == ord('q'):  # q to quit
                        raise StopIteration

                # Save results (image with detections)
                if save_img:
                    if dataset.mode == 'images':
                        cv2.imwrite(save_path, im0)
                        pass
                    else:
                        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))

    print('   All Done. (%.3fs)' % (time.time() - t0))


if __name__ == '__main__':
    """
        weights:训练的权重
        source:测试数据，可以是图片/视频路径，也可以是'0'(电脑自带摄像头),也可以是rtsp等视频流
        output:网络预测之后的图片/视频的保存路径
        img-size:网络输入图片大小
        conf-thres:置信度阈值
        iou-thres:做nms的iou阈值
        device:设置设备
        view-img:是否展示预测之后的图片/视频，默认False
        save-txt:是否将预测的框坐标以txt文件形式保存，默认False
        classes:设置只保留某一部分类别，形如0或者0 2 3
        agnostic-nms:进行nms是否将所有类别框一视同仁，默认False
        augment:推理的时候进行多尺度，翻转等操作(TTA)推理
        update:如果为True，则对所有模型进行strip_optimizer操作，去除pt文件中的优化器等信息，默认为False
    """
    parser = argparse.ArgumentParser()
    parser.add_argument('--weights', nargs='+', type=str, default='./runs/exp3/weights/last.pt', help='model.pt path(s)')
    parser.add_argument('--source', type=str, default='/data_all/data/DOTA/Ship_dota_v1.5_1024/val/images', help='source')  # file/folder, 0 for webcam
    parser.add_argument('--output', type=str, default='./detectionlx', help='output folder')  # output folder
    parser.add_argument('--img-size', type=int, default=1024, help='inference size (pixels)')
    parser.add_argument('--conf-thres', type=float, default=0.25, help='object confidence threshold')
    parser.add_argument('--iou-thres', type=float, default=0.4, help='IOU threshold for NMS')
    parser.add_argument('--device', default='0', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
    parser.add_argument('--view-img', action='store_true', help='display results')
    parser.add_argument('--save-txt', action='store_true', help='save results to *.txt')
    parser.add_argument('--classes', nargs='+', type=int, help='filter by class: --class 0, or --class 0 2 3')
    parser.add_argument('--agnostic-nms', action='store_true', default=False, help='class-agnostic NMS')
    parser.add_argument('--augment', action='store_true', help='augmented inference')
    parser.add_argument('--update', action='store_true', help='update all models')
    parser.add_argument('--kld', type=bool, default=True, help='use kld')
    opt = parser.parse_args()
    print(opt)

    with torch.no_grad():
        if opt.update:  # update all models (to fix SourceChangeWarning)
            for opt.weights in ['yolov5s.pt', 'yolov5m.pt', 'yolov5l.pt', 'yolov5x.pt']:
                detect(kld_flag=opt.kld)
                # 去除pt文件中的优化器等信息
                strip_optimizer(opt.weights)
        else:
            detect(kld_flag=opt.kld)