tensor

.idea/AI.iml

@@ -4,7 +4,7 @@
     <content url="file://$MODULE_DIR$">
       <excludeFolder url="file://$MODULE_DIR$/venv" />
     </content>
-    <orderEntry type="jdk" jdkName="Python 3.9 (torch1.7) (15)" jdkType="Python SDK" />
+    <orderEntry type="jdk" jdkName="Python 3.8 (yolov5)" jdkType="Python SDK" />
     <orderEntry type="sourceFolder" forTests="false" />
   </component>
 </module>

.idea/deployment.xml

@@ -2,6 +2,13 @@
 <project version="4">
   <component name="PublishConfigData" remoteFilesAllowedToDisappearOnAutoupload="false">
     <serverData>
+      <paths name="th@192.168.11.8:32178 password">
+        <serverdata>
+          <mappings>
+            <mapping local="$PROJECT_DIR$" web="/" />
+          </mappings>
+        </serverdata>
+      </paths>
       <paths name="thsw@192.168.10.11:22 password">
         <serverdata>
           <mappings>

.idea/misc.xml

@@ -1,4 +1,4 @@
 <?xml version="1.0" encoding="UTF-8"?>
 <project version="4">
-  <component name="ProjectRootManager" version="2" project-jdk-name="Python 3.9 (torch1.7) (15)" project-jdk-type="Python SDK" />
+  <component name="ProjectRootManager" version="2" project-jdk-name="Python 3.8 (yolov5)" project-jdk-type="Python SDK" />
 </project>

AI_example.py

@@ -6,6 +6,7 @@ import torch
 
 from DMPRUtils.DMPR_process import DMPR_process, plot_points
 from DMPRUtils.model.detector import DirectionalPointDetector
+from DMPRUtils.yolo_net import Model
 from DMPR_YOLO.jointUtil import dmpr_yolo
 from conf import config
 from models.experimental import attempt_load
@@ -16,12 +17,12 @@ 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'
+    device_ = '1'  ##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"
+    DMPRweights = "weights/urbanManagement/DMPR/dp_detector_299.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],
@@ -41,26 +42,29 @@ def main():
 
     # DMPR model
     args = config.get_parser_for_inference().parse_args()
-    DMPRmodel = DirectionalPointDetector(3, args.depth_factor, config.NUM_FEATURE_MAP_CHANNEL).to(device)
+    # DMPRmodel = DirectionalPointDetector(3, args.depth_factor, config.NUM_FEATURE_MAP_CHANNEL).to(device)
+    # DMPRmodel.load_state_dict(torch.load(DMPRweights))
+    DMPRmodel = Model(args.cfg, ch=3).to(device)
     DMPRmodel.load_state_dict(torch.load(DMPRweights))
 
     # 图像测试
-    # impth = 'images/input'
+    impth = 'images/input'
-    impth = 'images/debug'
+    # impth = 'images/debug'
-    # outpth = 'images/output'
+    # impth = '/home/thsw/WJ/zjc/AI_old/images/input_0'
-    outpth = 'images/debug_out'
+    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()
+        t_start = time.time()
         # yolo process
         det0 = yolo_process(img0, model, device, args, half)
         det0 = det0.cpu().detach().numpy()
-        # t_yolo = time.time()
+        t_yolo = time.time()
-        # print(f't_yolo. ({t_yolo - t_start:.3f}s)')
+        print(f't_yolo. ({t_yolo - t_start:.3f}s)')
 
         # plot所有box
         # for *xyxy, conf, cls in reversed(det0):
@@ -71,8 +75,8 @@ def main():
         det1 = DMPR_process(img0, DMPRmodel, device, args)
         det1 = det1.cpu().detach().numpy()
 
-        # t_dmpr = time.time()
+        t_dmpr = time.time()
-        # print(f't_dmpr. ({t_dmpr - t_yolo:.3f}s)')
+        print(f't_dmpr. ({t_dmpr - t_yolo:.3f}s)')
 
         # 绘制角点
         plot_points(img0, det1)
@@ -82,10 +86,10 @@ def main():
 
         # yolo joint DMPR
         cls = 0     #需要过滤的box类别
-        joint_det, dilate_box = dmpr_yolo(det1, det0, img0.shape, cls)
+        joint_det, dilate_box = dmpr_yolo(det1, det0, img0.shape, cls, args.scale_ratio)
 
-        # t_joint = time.time()
+        t_joint = time.time()
-        # print(f't_joint. ({t_joint - t_dmpr:.3f}s)')
+        print(f't_joint. ({t_joint - t_dmpr:.3f}s)')
 
         # t_end = time.time()
         # print(f'Done. ({t_end - t_start:.3f}s)')

DMPRUtils/DMPR_process.py

@@ -1,5 +1,6 @@
 import math
 import os
+import time
 from collections import namedtuple
 
 import cv2
@@ -132,7 +133,12 @@ def get_predicted_points2(prediction, thresh):
 
 def detect_marking_points(detector, image, thresh, device):
     """Given image read from opencv, return detected marking points."""
+    t1 = time.time()
+    torch.cuda.synchronize(device)
     prediction = detector(preprocess_image(image).to(device))
+    torch.cuda.synchronize(device)
+    t2 = time.time()
+    print(f'detector: {t2 - t1:.3f}s')
     return get_predicted_points2(prediction[0], thresh)
 
 def scale_coords2(img1_shape, coords, img0_shape, ratio_pad=None):

DMPR_YOLO/jointUtil.py

@@ -4,7 +4,7 @@ import numpy as np
 import torch
 
 
-def dmpr_yolo(dmpr_det, yolo_det, img_shape, cls:int):
+def dmpr_yolo(dmpr_det, yolo_det, img_shape, cls:int, scale_ratio):
 
     # 创建yolo_det_clone内容为x1, y1, x2, y2, conf, cls, unlabel (unlabel代表该类是否需要忽略，0：不忽略 其他：忽略)
     yolo_det_clone = yolo_det.copy()
@@ -22,8 +22,8 @@ def dmpr_yolo(dmpr_det, yolo_det, img_shape, cls:int):
     y_length = yolo_det[:, 3] - yolo_det[:, 1]     #y2-y1
 
     # x, y哪个方向差值大哪个方向膨胀的多
-    x_dilate_coefficient = ((x_length > y_length) + 1)*0.2
+    x_dilate_coefficient = ((x_length > y_length) + 1)*scale_ratio
-    y_dilate_coefficient = ((~(x_length > y_length)) + 1)*0.2
+    y_dilate_coefficient = ((~(x_length > y_length)) + 1)*scale_ratio
 
     # 膨胀
     new_yolo_det[:, 0] = np.round(yolo_det[:, 0] - x_dilate_coefficient * x_length).clip(0, img_shape[1])   #x1 膨胀
@@ -64,12 +64,24 @@ def dmpr_yolo(dmpr_det, yolo_det, img_shape, cls:int):
 
     direction1 = np.arctan2(y_c - y_p, x_c - x_p) / math.pi * 180
     direction2 = yolo_dmpr[..., 8] / math.pi * 180
+    # direction3 = (direction2 + 90) if (direction2 + 90) <= 180 else (direction2 - 270)
+    direction3 = direction2 + 90                    # L形角点另外一个方向
+    direction3[direction3 > 180] -= 360
     ang_diff = direction1 - direction2
+    ang_diff2 = direction1 - direction3
 
     # 判断膨胀后yolo框包含角点关系 & & 包含角点的时候计算水平框中心点与角点的角度关系
     # direction ∈ （-180， 180） 若角差大于180，需算补角
+    # T形角点比较一个方向，L形角点比较两个方向
     mask = (x_p >= x1) & (x_p <= x2) & (y_p >= y1) & (y_p <= y2) & \
-           (((ang_diff >= -90) & (ang_diff <= 90)) | ((ang_diff > 180) & ((360 - ang_diff) <= 90)) | (((ang_diff) < -180) & ((360 + ang_diff) <= 90)))
+           (((yolo_dmpr[..., 9] <= 0.5) &                          # T形角点情况
+                (((ang_diff >= -90) & (ang_diff <= 90)) | ((ang_diff > 180) & ((360 - ang_diff) <= 90)) |
+                    (((ang_diff) < -180) & ((360 + ang_diff) <= 90)))) |
+            ((yolo_dmpr[..., 9] > 0.5) &                            # L形角点情况
+                (((ang_diff >= -90) & (ang_diff <= 90)) | ((ang_diff > 180) & ((360 - ang_diff) <= 90)) |
+                    (((ang_diff) < -180) & ((360 + ang_diff) <= 90))) &
+                (((ang_diff2 >= -90) & (ang_diff2 <= 90)) | ((ang_diff2 > 180) & ((360 - ang_diff2) <= 90)) |
+                    (((ang_diff2) < -180) & ((360 + ang_diff2) <= 90)))))
 
     res = np.sum(mask, axis=1)
 

conf/config.py

@@ -5,11 +5,14 @@ import argparse
 # 5: sin(direction)
 NUM_FEATURE_MAP_CHANNEL = 6
 
-
 def add_common_arguments(parser):
     """Add common arguments for training and inference."""
     parser.add_argument('--detector_weights', default=r'E:\pycharmProject\DMPR-PS\weights\dp_detector_499.pth',
                         help="The weights of pretrained detector.")
+    parser.add_argument('--cfg', type=str, default='models/yolov5s.yaml',
+                        help='model.yaml path')
+    parser.add_argument('--hyp', type=str, default='conf/hyp.scratch.yaml',
+                        help='hyperparameters path')
     parser.add_argument('--depth_factor', type=int, default=32,
                         help="Depth factor.")
     parser.add_argument('--disable_cuda', action='store_true',
@@ -41,6 +44,8 @@ def get_parser_for_inference():
                         help='object confidence threshold')
     parser.add_argument('--iou-thres', type=float, default=0.45,
                         help='IOU threshold for NMS')
+    parser.add_argument('--scale-ratio', type=float, default=0.5,
+                        help='detected box scale ratio')
     parser.add_argument('--ovlap-thres', type=float, default=0.6, help='overlap threshold for OBS')
     parser.add_argument('--agnostic-nms', action='store_true',
                         help='class-agnostic NMS')