urban_management/AI_example.py

import os
import time

import cv2
import torch

from DMPRUtils.DMPR_process import DMPR_process, plot_points
from DMPRUtils.model.detector import DirectionalPointDetector
from DMPR_YOLO.jointUtil import dmpr_yolo
from conf import config
from models.experimental import attempt_load
from models.yolo_process import yolo_process
from utils.plots import plot_one_box
from utils.torch_utils import select_device


def main():
    ##预先设置的参数
    device_ = '0'  ##选定模型，可选 cpu,'0','1'

    ##以下参数目前不可改
    Detweights = 'weights/urbanManagement/yolo/best.pt'
    seg_nclass = 2
    DMPRweights = "weights/urbanManagement/DMPR/dp_detector_499.pth"
    conf_thres, iou_thres, classes = 0.25, 0.45, 3
    labelnames = "weights/yolov5/class5/labelnames.json"
    rainbows = [[0, 0, 255], [0, 255, 0], [255, 0, 0], [255, 0, 255], [255, 255, 0], [255, 129, 0], [255, 0, 127],
                [127, 255, 0], [0, 255, 127], [0, 127, 255], [127, 0, 255], [255, 127, 255], [255, 255, 127],
                [127, 255, 255], [0, 255, 255], [255, 127, 255], [127, 255, 255], [0, 127, 0], [0, 0, 127],
                [0, 255, 255]]
    allowedList = [0, 1, 2, 3]

    ##加载模型，准备好显示字符
    device = select_device(device_)

    half = device.type != 'cpu'  # half precision only supported on CUDA
    # yolov5 model
    model = attempt_load(Detweights, map_location=device)
    if half:
        model.half()

    # DMPR model
    args = config.get_parser_for_inference().parse_args()
    DMPRmodel = DirectionalPointDetector(3, args.depth_factor, config.NUM_FEATURE_MAP_CHANNEL).to(device)
    DMPRmodel.load_state_dict(torch.load(DMPRweights))

    # 图像测试
    # impth = 'images/input'
    impth = 'images/debug'
    # outpth = 'images/output'
    outpth = 'images/debug_out'
    folders = os.listdir(impth)
    for file in folders:
        imgpath = os.path.join(impth, file)
        img0 = cv2.imread(imgpath)
        assert img0 is not None, 'Image Not Found ' + imgpath

        # t_start = time.time()
        # yolo process
        det0 = yolo_process(img0, model, device, args, half)
        det0 = det0.cpu().detach().numpy()
        # t_yolo = time.time()
        # print(f't_yolo. ({t_yolo - t_start:.3f}s)')

        # plot所有box
        # for *xyxy, conf, cls in reversed(det0):
        #     label = f'{int(cls)} {conf:.2f}'
        #     plot_one_box(xyxy, img0, label=label, color=rainbows[int(cls)], line_thickness=2)

        # DMPR process
        det1 = DMPR_process(img0, DMPRmodel, device, args)
        det1 = det1.cpu().detach().numpy()

        # t_dmpr = time.time()
        # print(f't_dmpr. ({t_dmpr - t_yolo:.3f}s)')

        # 绘制角点
        plot_points(img0, det1)

        # save
        # cv2.imwrite(file, img0)

        # yolo joint DMPR
        cls = 0     #需要过滤的box类别
        joint_det, dilate_box = dmpr_yolo(det1, det0, img0.shape, cls)

        # t_joint = time.time()
        # print(f't_joint. ({t_joint - t_dmpr:.3f}s)')

        # t_end = time.time()
        # print(f'Done. ({t_end - t_start:.3f}s)')
        # 绘制膨胀box
        for *xyxy, flag in dilate_box:
            plot_one_box(xyxy, img0, color=rainbows[int(cls)], line_thickness=2)
        #
        # # 绘制删除满足 在膨胀框内 && 角度差小于90度 的box
        for *xyxy, conf, cls, flag in reversed(joint_det):
            if flag == 0:
                # label = f'{int(cls)} {conf:.2f}'
                label = None
                plot_one_box(xyxy, img0, label=label, color=rainbows[int(cls)], line_thickness=2)

        # save
        save_path = os.path.join(outpth, file)
        cv2.imwrite(save_path, img0)



if __name__ == '__main__':
    main()