decode_fn = Cityscapes.decode_target
gpu_id:只有一个gpu的话输入0即可(注意是字符串类型)
input：图像输入路径（可以是图片也可以是图片所在文件夹）
save_val_results_to:图像输出路径
ckpt==best_deeplabv3plus_mobilenet_cityscapes_os16.pth

#!/usr/bin/env python
from __future__ import print_function
from tkinter import image_names
import roslib
import sys
from torch.utils.data import dataset
from tqdm import tqdm
import network
import utils
import os
import random
import argparse
import numpy as np
 
from torch.utils import data
from datasets import VOCSegmentation, Cityscapes, cityscapes
from torchvision import transforms as T
from metrics import StreamSegMetrics
import cv2
import torch
import torch.nn as nn
 
from PIL import Image
import matplotlib
import matplotlib.pyplot as plt
from glob import glob
 
import roslib
#roslib.load_manifest('my_package')
import sys
import rospy
import cv2
from std_msgs.msg import String
from sensor_msgs.msg import Image
from cv_bridge import CvBridge, CvBridgeError
 
 
# def get_argparser():
    
#     parser = argparse.ArgumentParser()
 
#     # Datset Options
#     parser.add_argument("--input", type=str, required=True,
#                         help="path to a single image or image directory")
#     parser.add_argument("--dataset", type=str, default='voc',
#                         choices=['voc', 'cityscapes'], help='Name of training set')
 
#     # Deeplab Options
#     available_models = sorted(name for name in network.modeling.__dict__ if name.islower() and \
#                               not (name.startswith("__") or name.startswith('_')) and callable(
#                               network.modeling.__dict__[name])
#                               )
 
#     parser.add_argument("--model", type=str, default='deeplabv3plus_mobilenet',
#                         choices=available_models, help='model name')
#     parser.add_argument("--separable_conv", action='store_true', default=False,
#                         help="apply separable conv to decoder and aspp")
#     parser.add_argument("--output_stride", type=int, default=16, choices=[8, 16])
 
#     # Train Options
#     parser.add_argument("--save_val_results_to", default=None,
#                         help="save segmentation results to the specified dir")
 
#     parser.add_argument("--crop_val", action='store_true', default=False,
#                         help='crop validation (default: False)')
#     parser.add_argument("--val_batch_size", type=int, default=4,
#                         help='batch size for validation (default: 4)')
#     parser.add_argument("--crop_size", type=int, default=513)
 
    
#     parser.add_argument("--ckpt", default=None, type=str,
#                         help="resume from checkpoint")
#     parser.add_argument("--gpu_id", type=str, default='0',
#                         help="GPU ID")
#     return parser
 
if __name__ == '__main__':
    print("somthing......")
    rospy.init_node('kitti',anonymous=True)
    print("somthing......")
    bridge=CvBridge()
    image_pub = rospy.Publisher("image_orioi",Image,queue_size=10)
    image_sematic_pub=rospy.Publisher("sematic_image",Image,queue_size=10)
    #opts = get_argparser().parse_args()
    # if opts.dataset.lower() == 'voc':
    #     opts.num_classes = 21
    #     decode_fn = VOCSegmentation.decode_target
    # elif opts.dataset.lower() == 'cityscapes':
    #     opts.num_classes = 19
    #     #1
    #     decode_fn = Cityscapes.decode_target
    decode_fn = Cityscapes.decode_target
    print("somthing2......")
    #2 defalut=0
    gpu_id="0"
    #os.environ['CUDA_VISIBLE_DEVICES'] = opts.gpu_id
    os.environ['CUDA_VISIBLE_DEVICES'] = gpu_id
    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
    print("Device: %s" % device)
    
    # Setup dataloader
    image_files = []
    #3 input  /home/du/work/study/ros/aloam_ws/src/deeplab/src/aachen/aachen_000000_000019_leftImg8bit.png
    #input="/home/du/work/study/ros/aloam_ws/src/deeplab/src/aachen"
    input="/home/du/work/study/ros/aloam_ws/cowtransfer/2011_09_26_drive_0002_sync/2011_09_26/2011_09_26_drive_0002_sync/image_03/data"
    if os.path.isdir(input):
        
        for file_name in os.listdir(input):
            data_collect=input+'/'+file_name
            image_files.append(data_collect)
    #5 input  /home/du/work/study/ros/aloam_ws/src/deeplab/src/aachen/aachen_000000_000019_leftImg8bit.png
    elif os.path.isfile(input):
        image_files.append(input)
    
    # Set up model (all models are 'constructed at network.modeling)
    #6  num_classes == 19 output_stride==16 defalut
    model = network.modeling.__dict__["deeplabv3plus_mobilenet"](num_classes=19, output_stride=16)
    # if opts.separable_conv and 'plus' in "deeplabv3plus_mobilenet":
    #     network.convert_to_separable_conv(model.classifier)
    print("somthing3......")
    utils.set_bn_momentum(model.backbone, momentum=0.01)
    print("somthing4......")
    #7 ckpt==best_deeplabv3plus_mobilenet_cityscapes_os16.pth
    if "/home/du/work/study/ros/aloam_ws/src/deeplab/src/best_deeplabv3plus_mobilenet_cityscapes_os16.pth" is not None and os.path.isfile("/home/du/work/study/ros/aloam_ws/src/deeplab/src/best_deeplabv3plus_mobilenet_cityscapes_os16.pth"):
        # https://github.com/VainF/DeepLabV3Plus-Pytorch/issues/8#issuecomment-605601402, @PytaichukBohdan
        checkpoint = torch.load("/home/du/work/study/ros/aloam_ws/src/deeplab/src/best_deeplabv3plus_mobilenet_cityscapes_os16.pth", map_location=torch.device('cpu'))
        model.load_state_dict(checkpoint["model_state"])
        model = nn.DataParallel(model)
        model.to(device)
        print("Resume model from %s" % "/home/du/work/study/ros/aloam_ws/src/deeplab/src/best_deeplabv3plus_mobilenet_cityscapes_os16.pth")
        del checkpoint
    else:
        print("[!] Retrain")
        model = nn.DataParallel(model)
        model.to(device)
 
    #denorm = utils.Denormalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])  # denormalization for ori images
    # defalut == false
    # if opts.crop_val:
    #     transform = T.Compose([
    #             T.Resize(opts.crop_size),
    #             T.CenterCrop(opts.crop_size),
    #             T.ToTensor(),
    #             T.Normalize(mean=[0.485, 0.456, 0.406],
    #                             std=[0.229, 0.224, 0.225]),
    #         ])
    # else:
    transform = T.Compose([
            T.ToTensor(),
            T.Normalize(mean=[0.485, 0.456, 0.406],
                            std=[0.229, 0.224, 0.225]),
        ])
    # if opts.save_val_results_to is not None:
    #     #os.makedirs(opts.save_val_results_to, exist_ok=True)
    #     os.makedirs(opts.save_val_results_to)
    with torch.no_grad():
        model = model.eval()
        for img_path in tqdm(image_files):
            print("root is")
            print(img_path)
            ext = os.path.basename(img_path).split('.')[-1]
            img_name = os.path.basename(img_path)[:-len(ext)-1]
            print(img_name)
            #img = Image.open(img_path).convert('RGB')
            #opencv
            img_bgr=cv2.imread(img_path)
            #BGR->RGB
            img = cv2.cvtColor(img_bgr, cv2.COLOR_BGR2RGB)
            imgcv_msg=bridge.cv2_to_imgmsg(img,"bgr8")
            image_pub.publish(imgcv_msg)
            #cv->tensor
            img = transform(img).unsqueeze(0) # To tensor of NCHW
            img = img.to(device)
            #predict
            pred = model(img).max(1)[1].cpu().numpy()[0] # HW
            colorized_preds = decode_fn(pred).astype('uint8')
            #numpy->opencv
            img_cv = cv2.cvtColor(colorized_preds,cv2.COLOR_RGB2BGR)
            #opencv->rosmsg
            img_msg=bridge.cv2_to_imgmsg(img_cv,"bgr8")
 
            image_sematic_pub.publish(img_msg)
            #print("publish success")
 
            #numpy->Image
            #colorized_preds = Image.fromarray(colorized_preds)
            
                #colorized_preds.save(os.path.join(opts.save_val_results_to, img_name+'.png'))
            cv2.imwrite(os.path.join("/home/du/work/study/ros/aloam_ws/src/deeplab/src/test_results", img_name+'.png'),img_cv)
            print(os.path.join("/home/du/work/study/ros/aloam_ws/src/deeplab/src/test_results", img_name+'.png'))
            # rospy.init_node('sematic_img', anonymous=True)
            # try:
            #     rospy.spin()
            # except KeyboardInterrupt:
            #     print("Shutting down")
 

打开anaconda的py2.7环境
conda activate py2.7
再source一下
source devel/setup.bash
运行程序
rosrun deeplab predict.py

#!/usr/bin/env python
from __future__ import print_function
from tkinter import image_names
import roslib
import sys
from torch.utils.data import dataset
from tqdm import tqdm
import network
import utils
import os
import random
import argparse
import numpy as np
 
from torch.utils import data
from datasets import VOCSegmentation, Cityscapes, cityscapes
from torchvision import transforms as T
from metrics import StreamSegMetrics
import cv2
import torch
import torch.nn as nn
 
from PIL import Image
import matplotlib
import matplotlib.pyplot as plt
from glob import glob
 
import roslib
#roslib.load_manifest('my_package')
import sys
import rospy
import cv2
from std_msgs.msg import String
from sensor_msgs.msg import Image
from cv_bridge import CvBridge, CvBridgeError
 
 
class sematicCloud:
    def __init__(self) :
        self.image_pub = rospy.Publisher("image_orioi",Image,queue_size=10)
        self.sematic_img=rospy.Publisher("sematic_image",Image,queue_size=10)
        self.image_sub = rospy.Subscriber("/kitti/camera_color_left/image_raw",Image,self.callback)
        self.bridge = CvBridge()
        
        self.decode_fn = Cityscapes.decode_target
        self.gpu_id="0"
        os.environ['CUDA_VISIBLE_DEVICES'] = self.gpu_id
        self.device=torch.device('cuda' if torch.cuda.is_available() else 'cpu')
        self.model=network.modeling.__dict__["deeplabv3plus_mobilenet"](num_classes=19, output_stride=16)
        self.checkpoint = torch.load("/home/du/work/study/ros/aloam_ws/src/deeplab/src/best_deeplabv3plus_mobilenet_cityscapes_os16.pth", map_location=torch.device('cpu'))
        self.model.load_state_dict(self.checkpoint["model_state"])
        self.model = nn.DataParallel(self.model)
        self.model=self.model.to(self.device).eval()
        self.model.to(self.device)
        print("Resume model from %s" % "/home/du/work/study/ros/aloam_ws/src/deeplab/src/best_deeplabv3plus_mobilenet_cityscapes_os16.pth")
        del self.checkpoint
 
        self.transform = T.Compose([
            T.ToTensor(),
            T.Normalize(mean=[0.485, 0.456, 0.406],
                            std=[0.229, 0.224, 0.225]),
        ])
    def callback(self,data):
        print("hello world")
        try:
            #rosmsg->cv
            img_bgr = self.bridge.imgmsg_to_cv2(data, "bgr8")
        except CvBridgeError as e:
            print(e)
 
        img = cv2.cvtColor(img_bgr, cv2.COLOR_BGR2RGB)
 
        imgcv_msg=self.bridge.cv2_to_imgmsg(img,"bgr8")
 
        self.image_pub.publish(imgcv_msg)
        #cv->tensor
        img = self.transform(img).unsqueeze(0) # To tensor of NCHW
 
        img = img.to(self.device)
 
        #predict
        pred = self.model(img).max(1)[1].cpu().numpy()[0] # HW
 
        colorized_preds = self.decode_fn(pred).astype('uint8')
     
        #numpy->opencv
        img_cv = cv2.cvtColor(colorized_preds,cv2.COLOR_RGB2BGR)
        #opencv->rosmsg
        img_msg=self.bridge.cv2_to_imgmsg(img_cv,"bgr8")
        img_msg.header.stamp=data.header.stamp
        self.sematic_img.publish(img_msg)
        # cv2.imwrite(os.path.join("/home/du/work/study/ros/aloam_ws/src/deeplab/src/test_results", img_name+'.png'),img_cv)
        # print(os.path.join("/home/du/work/study/ros/aloam_ws/src/deeplab/src/test_results", img_name+'.png'))
 
 
def main(args):
  sc = sematicCloud()
  rospy.init_node('image_converter', anonymous=True)
  try:
    rospy.spin()
  except KeyboardInterrupt:
    print("Shutting down")
  cv2.destroyAllWindows()
 
if __name__ == '__main__':
    main(sys.argv)

# -*- coding: cp936 -*-
import cv2
import numpy as np
 
def nothing(x):
  pass
 
img = cv2.imread('/home/du/work/study/opencv/Python_OpenCV_yuanma/Lesson31/yuyi.png')
 
cv2.namedWindow('hsv_debug')
 
cv2.createTrackbar('h_min','hsv_debug',0,180,nothing)
cv2.createTrackbar('s_min','hsv_debug',0,255,nothing)
cv2.createTrackbar('v_min','hsv_debug',0,255,nothing)
 
cv2.createTrackbar('h_max','hsv_debug',0,180,nothing)
cv2.createTrackbar('s_max','hsv_debug',0,255,nothing)
cv2.createTrackbar('v_max','hsv_debug',0,255,nothing)
 
hsv_img=cv2.cvtColor(img,cv2.COLOR_BGR2HSV)
 
while(1): 
  h_min=cv2.getTrackbarPos('h_min','hsv_debug')
  s_min=cv2.getTrackbarPos('s_min','hsv_debug')
  v_min=cv2.getTrackbarPos('v_min','hsv_debug')
 
  h_max=cv2.getTrackbarPos('h_max','hsv_debug')
  s_max=cv2.getTrackbarPos('s_max','hsv_debug')
  v_max=cv2.getTrackbarPos('v_max','hsv_debug')
 
  lower_green = np.array([h_min,s_min,v_min])
  upper_green = np.array([h_max,s_max,v_max])
 
  mask = cv2.inRange(hsv_img,lower_green,upper_green)
 
  cv2.imshow('hsv_debug',mask)
  k=cv2.waitKey(1)&0xFF
  if k==27:
    break
 
cv2.destroyAllWindows()

def dectImage(self,img_orioi,img_sematic):
      img_orioi_copy=img_orioi
      img_sematic_copy=img_sematic
      hsv_image_sematic=cv2.cvtColor(img_sematic_copy,cv2.COLOR_BGR2HSV)
 
      lower_blue=np.array([100,43,46])
      upper_blue=np.array([124,255,153])
      lower_red = np.array([0,122,120])
      upper_red = np.array([10,255,255])
 
      mask_red = cv2.inRange(hsv_image_sematic,lower_red,upper_red)
      mask_red = cv2.medianBlur(mask_red, 5)
      mask_red,contours_red,hierarchy = cv2.findContours(mask_red, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)
 
      mask_blue = cv2.inRange(hsv_image_sematic,lower_blue,upper_blue)
      mask_blue = cv2.medianBlur(mask_blue, 5)
      mask_blue,contours_blue,hierarchy = cv2.findContours(mask_blue, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)
 
 
      if contours_red:
        for cnt in contours_red:
          (x, y, w, h) = cv2.boundingRect(cnt)
          cv2.rectangle(img_orioi_copy,(x,y),(x+w,y+h),(0,255,255),2)
          cv2.putText(img_orioi_copy,"Person",(x,y),font,0.7,(0,255,0),2)
 
          #cv->rosmsg
          imgcv_dect=self.bridge.cv2_to_imgmsg(img_orioi_copy,"bgr8")
          self.img_dect_pub.publish(imgcv_dect)
          print("dect pub seccess")
 
      elif contours_blue:
        for cnt in contours_blue:
          (x, y, w, h) = cv2.boundingRect(cnt)
          cv2.rectangle(img_orioi_copy,(x,y),(x+w,y+h),(0,0,255),2)
          cv2.putText(img_orioi_copy,"Car",(x,y),font,0.7,(0,0,255),2)
 
          #cv->rosmsg
          imgcv_dect=self.bridge.cv2_to_imgmsg(img_orioi_copy,"bgr8")
          self.img_dect_pub.publish(imgcv_dect)
          print("dect pub seccess")
      else:
        imgcv_dect=self.bridge.cv2_to_imgmsg(img_orioi_copy,"bgr8")
        self.img_dect_pub.publish(imgcv_dect)
        print("dect pub seccess")