\u76ee\u5f55<\/p>\n
\u4e00\u3001DNN\u98ce\u683c\u8fc1\u79fb：\u539f\u7406\u4e0e\u4ee3\u7801\u5b9e\u73b0<\/p>\n
1. \u6838\u5fc3\u539f\u7406<\/p>\n
2. \u5b9e\u6218\u4ee3\u7801\u5b9e\u73b0<\/p>\n
3. \u5173\u952e\u51fd\u6570\u89e3\u6790<\/p>\n
\u4e8c\u3001CSRT\u7269\u4f53\u8ffd\u8e2a：\u539f\u7406\u4e0e\u4ee3\u7801\u5b9e\u73b0<\/p>\n
1. \u6838\u5fc3\u539f\u7406<\/p>\n
2. \u5b9e\u6218\u4ee3\u7801\u5b9e\u73b0<\/p>\n
3. \u5173\u952e\u51fd\u6570\u89e3\u6790<\/p>\n
\u4e09\u3001\u4e24\u8005\u6574\u5408：\u5b9e\u65f6\u98ce\u683c\u8fc1\u79fb+\u7269\u4f53\u8ffd\u8e2a<\/p>\n
1. \u6574\u5408\u6838\u5fc3\u903b\u8f91<\/p>\n
2. \u6574\u5408\u540e\u5b8c\u6574\u4ee3\u7801<\/p>\n
3. \u6574\u5408\u5173\u952e\u4f18\u5316\u70b9<\/p>\n
\u56db\u3001\u5b9e\u6218\u6ce8\u610f\u4e8b\u9879<\/p>\n
\u5728\u8ba1\u7b97\u673a\u89c6\u89c9\u5b9e\u6218\u4e2d，\u5c06\u98ce\u683c\u8fc1\u79fb\u4e0e\u7269\u4f53\u8ffd\u8e2a\u7ed3\u5408\u53ef\u5b9e\u73b0\u66f4\u5177\u89c6\u89c9\u51b2\u51fb\u529b\u7684\u6548\u679c\u3002\u672c\u6587\u57fa\u4e8eOpenCV，\u4ece\u539f\u7406\u5230\u4ee3\u7801\u9010\u6a21\u5757\u62c6\u89e3，\u6700\u7ec8\u5b9e\u73b0\u201c\u5b9e\u65f6\u98ce\u683c\u8fc1\u79fb+\u76ee\u6807\u8ffd\u8e2a\u201d\u7684\u878d\u5408\u65b9\u6848，\u5168\u7a0b\u805a\u7126\u5b9e\u64cd，\u4e0d\u5806\u780c\u5197\u4f59\u7406\u8bba\u3002<\/p>\n
DNN（Deep Neural Network，\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc）\u662f\u6df1\u5ea6\u5b66\u4e60\u7684\u6838\u5fc3\u8f7d\u4f53，\u901a\u8fc7\u591a\u5c42\u795e\u7ecf\u5143\u6a21\u62df\u4eba\u8111\u7ed3\u6784，\u5b9e\u73b0\u4ece\u6570\u636e\u4e2d\u81ea\u52a8\u63d0\u53d6\u7279\u5f81\u3001\u5b8c\u6210\u590d\u6742\u4efb\u52a1（\u5982\u56fe\u50cf\u8bc6\u522b\u3001\u98ce\u683c\u8fc1\u79fb）\u3002\u4e0e\u4f20\u7edf\u6d45\u5c42\u7f51\u7edc\u76f8\u6bd4，DNN\u51ed\u501f\u6df1\u5c42\u7ed3\u6784\u53ef\u6355\u6349\u66f4\u62bd\u8c61\u7684\u7279\u5f81，\u5728\u8ba1\u7b97\u673a\u89c6\u89c9\u9886\u57df\u4e2d，\u9884\u8bad\u7ec3DNN\u6a21\u578b\u80fd\u8df3\u8fc7\u7e41\u7410\u7684\u7279\u5f81\u5de5\u7a0b，\u76f4\u63a5\u7528\u4e8e\u63a8\u7406\u90e8\u7f72，\u672c\u6587\u98ce\u683c\u8fc1\u79fb\u6b63\u662f\u57fa\u4e8e\u9884\u8bad\u7ec3DNN\u6a21\u578b\u5feb\u901f\u5b9e\u73b0\u3002<\/p>\n
\u57fa\u4e8e\u9884\u8bad\u7ec3DNN\u6a21\u578b\u7684\u98ce\u683c\u8fc1\u79fb，\u672c\u8d28\u662f\u901a\u8fc7\u795e\u7ecf\u7f51\u7edc\u63d0\u53d6\u8f93\u5165\u56fe\u50cf\u7684\u5185\u5bb9\u7279\u5f81\u4e0e\u98ce\u683c\u56fe\u50cf\u7684\u98ce\u683c\u7279\u5f81，\u518d\u878d\u5408\u751f\u6210\u65b0\u56fe\u50cf\u3002\u672c\u6587\u76f4\u63a5\u4f7f\u7528\u9884\u8bad\u7ec3\u7684`.t7`\u6a21\u578b（\u57fa\u4e8eTorch\u6846\u67b6\u8bad\u7ec3），\u8df3\u8fc7\u8bad\u7ec3\u73af\u8282，\u4e13\u6ce8\u63a8\u7406\u90e8\u7f72，\u517c\u987e\u901f\u5ea6\u4e0e\u6548\u679c\u3002<\/p>\n
\u6838\u5fc3\u6d41\u7a0b：\u8bfb\u53d6\u56fe\u50cf\u2192\u751f\u6210\u7b26\u5408\u6a21\u578b\u8f93\u5165\u8981\u6c42\u7684Blob\u2192\u6a21\u578b\u524d\u5411\u4f20\u64ad\u2192\u7ed3\u679c\u540e\u5904\u7406（\u7ef4\u5ea6\u8f6c\u6362\u3001\u5f52\u4e00\u5316）\u2192\u8f93\u51fa\u98ce\u683c\u5316\u56fe\u50cf\u3002<\/p>\n
\u4f9d\u8d56\u5e93：`opencv-python`\u3001`numpy`，\u9700\u63d0\u524d\u51c6\u59074\u4e2a\u9884\u8bad\u7ec3\u6a21\u578b（\u653e\u7f6e\u4e8e`model`\u6587\u4ef6\u5939）：`composition_vii.t7`\u3001`candy.t7`\u3001`udnie.t7`\u3001`feathers.t7`\u3002<\/p>\n
import cv2
\nimport numpy as np<\/p>\n
# 1. \u52a0\u8f7d\u9884\u8bad\u7ec3\u6a21\u578b（\u4ec5\u52a0\u8f7d\u4e00\u6b21，\u63d0\u5347\u6548\u7387）
\nmodels = {
\n "left_top": cv2.dnn.readNet("model\/composition_vii.t7"),
\n "right_top": cv2.dnn.readNet("model\/candy.t7"),
\n "left_bottom": cv2.dnn.readNet("model\/udnie.t7"),
\n "right_bottom": cv2.dnn.readNet("model\/feathers.t7")
\n}<\/p>\n
# 2. \u98ce\u683c\u8fc1\u79fb\u6838\u5fc3\u51fd\u6570
\ndef style_transfer(img, net):
\n h, w = img.shape[:2]
\n # \u751f\u6210\u6a21\u578b\u8f93\u5165Blob：\u56fe\u50cf\u5f52\u4e00\u5316\u3001\u5c3a\u5bf8\u8c03\u6574\u3001\u5747\u503c\u51cf\u6cd5
\n blob = cv2.dnn.blobFromImage(
\n img, 1.0, (w, h), (103.939, 116.779, 123.680),
\n swapRB=True, crop=False
\n )
\n # \u8bbe\u7f6e\u6a21\u578b\u8f93\u5165
\n net.setInput(blob)
\n # \u524d\u5411\u4f20\u64ad\u63a8\u7406\u98ce\u683c\u5316\u7ed3\u679c
\n preds = net.forward()<\/p>\n
# \u7ed3\u679c\u540e\u5904\u7406：\u7ef4\u5ea6\u8f6c\u6362（1,3,H,W）\u2192（H,W,3）
\n preds = preds.reshape(3, h, w)
\n preds = np.transpose(preds, (1, 2, 0))
\n # \u5f52\u4e00\u5316\u52300-255\u5e76\u8f6c\u4e3auint8（OpenCV\u517c\u5bb9\u683c\u5f0f）
\n preds = (preds – preds.min()) \/ (preds.max() – preds.min() + 1e-8)
\n preds = (preds * 255).astype(np.uint8)
\n # \u786e\u4fdd3\u901a\u9053（\u517c\u5bb9\u7070\u5ea6\u56fe\u8f93\u5165）
\n if preds.shape[2] != 3:
\n preds = cv2.cvtColor(preds, cv2.COLOR_GRAY2BGR)
\n return preds<\/p>\n
# 3. \u6444\u50cf\u5934\u5b9e\u65f6\u98ce\u683c\u8fc1\u79fb（\u56db\u533a\u57df\u5206\u98ce\u683c）
\ncap = cv2.VideoCapture(0)
\nwhile True:
\n ret, frame = cap.read()
\n if not ret:
\n break
\n h, w = frame.shape[:2]
\n h_half, w_half = h\/\/2, w\/\/2<\/p>\n
# \u5206\u5272\u56fe\u50cf\u4e3a\u56db\u533a\u57df，\u5206\u522b\u98ce\u683c\u8fc1\u79fb
\n tl = style_transfer(frame[:h_half, :w_half], models["left_top"])
\n tr = style_transfer(frame[:h_half, w_half:], models["right_top"])
\n bl = style_transfer(frame[h_half:, :w_half], models["left_bottom"])
\n br = style_transfer(frame[h_half:, w_half:], models["right_bottom"])<\/p>\n
# \u62fc\u63a5\u7ed3\u679c\u5e76\u663e\u793a
\n top_row = np.hstack((tl, tr))
\n bottom_row = np.hstack((bl, br))
\n merged = np.vstack((top_row, bottom_row))
\n cv2.imshow("Style Transfer", merged)<\/p>\n
if cv2.waitKey(1) == 27:
\n break
\ncap.release()
\ncv2.destroyAllWindows() <\/p>\n
cv2.dnn.readNet()：\u52a0\u8f7d\u9884\u8bad\u7ec3DNN\u6a21\u578b，\u652f\u6301`.t7`\u3001`.pb`\u7b49\u591a\u79cd\u683c\u5f0f，\u6b64\u5904\u7528\u4e8e\u8bfb\u53d6\u98ce\u683c\u8fc1\u79fb\u6a21\u578b\u3002<\/p>\n<\/li>\n
cv2.dnn.blobFromImage()：\u5c06\u56fe\u50cf\u8f6c\u6362\u4e3a\u6a21\u578b\u53ef\u63a5\u53d7\u7684Blob\u683c\u5f0f，\u6838\u5fc3\u53c2\u6570\u9700\u5339\u914d\u6a21\u578b\u8bad\u7ec3\u65f6\u7684\u914d\u7f6e（\u5747\u503c\u3001\u901a\u9053\u4ea4\u6362\u7b49），\u5426\u5219\u4f1a\u5bfc\u81f4\u98ce\u683c\u5f02\u5e38\u3002<\/p>\n<\/li>\n
net.setInput()：\u4e3aDNN\u6a21\u578b\u8bbe\u7f6e\u8f93\u5165\u6570\u636e（Blob\u683c\u5f0f），\u662f\u63a8\u7406\u524d\u7684\u5fc5\u8981\u6b65\u9aa4\u3002<\/p>\n<\/li>\n
net.forward()：\u6267\u884c\u6a21\u578b\u524d\u5411\u4f20\u64ad，\u8fd4\u56de\u98ce\u683c\u5316\u540e\u7684\u7279\u5f81\u56fe，\u9700\u540e\u7eed\u7ef4\u5ea6\u8f6c\u6362\u624d\u80fd\u4f5c\u4e3a\u56fe\u50cf\u663e\u793a\u3002<\/p>\n<\/li>\n
np.transpose()：\u8c03\u6574\u6570\u7ec4\u7ef4\u5ea6，\u5c06\u6a21\u578b\u8f93\u51fa\u7684（\u901a\u9053\u6570，\u9ad8\u5ea6，\u5bbd\u5ea6）\u8f6c\u4e3aOpenCV\u652f\u6301\u7684（\u9ad8\u5ea6，\u5bbd\u5ea6，\u901a\u9053\u6570）\u3002<\/p>\n<\/li>\n<\/ul>\n
CSRT（Channel and Spatial Reliability Tracking）\u662fOpenCV\u5185\u7f6e\u7684\u9ad8\u7cbe\u5ea6\u8ffd\u8e2a\u7b97\u6cd5，\u7ed3\u5408\u4e86\u901a\u9053\u53ef\u9760\u6027\u548c\u7a7a\u95f4\u53ef\u9760\u6027\u8bc4\u4f30，\u80fd\u9002\u5e94\u5149\u7167\u53d8\u5316\u3001\u76ee\u6807\u7f29\u653e\u7b49\u573a\u666f，\u7cbe\u5ea6\u4f18\u4e8eKCF\u3001BOOSTING\u7b49\u7b97\u6cd5，\u9002\u5408\u5b9e\u65f6\u573a\u666f\u3002<\/p>\n
\u6838\u5fc3\u6d41\u7a0b：\u521d\u59cb\u5316\u8ffd\u8e2a\u5668\u2192\u624b\u52a8\u9009\u62e9ROI（\u611f\u5174\u8da3\u533a\u57df）\u2192\u9010\u5e27\u66f4\u65b0\u8ffd\u8e2a\u5668\u2192\u83b7\u53d6\u76ee\u6807\u5750\u6807\u5e76\u7ed8\u5236\u3002<\/p>\n
import cv2<\/p>\n
# 1. \u521d\u59cb\u5316CSRT\u8ffd\u8e2a\u5668
\ntracker = cv2.TrackerCSRT_create()
\ntracking = False # \u8ffd\u8e2a\u72b6\u6001\u6807\u5fd7
\ncap = cv2.VideoCapture(0)<\/p>\n
while True:
\n ret, frame = cap.read()
\n if not ret:
\n break<\/p>\n
# 2. \u6309\u952e\u63a7\u5236：\u6309's'\u9009\u62e9ROI\u5f00\u59cb\u8ffd\u8e2a
\n key = cv2.waitKey(1) & 0xFF
\n if key == ord('s'):
\n tracking = True
\n # \u9009\u62e9ROI（\u611f\u5174\u8da3\u533a\u57df），\u8fd4\u56de（x,y,w,h）
\n roi = cv2.selectROI("CSRT Tracking", frame, showCrosshair=False)
\n # \u521d\u59cb\u5316\u8ffd\u8e2a\u5668（\u7ed1\u5b9a\u5f53\u524d\u5e27\u548cROI）
\n tracker.init(frame, roi)
\n elif key == 27:
\n break<\/p>\n
# 3. \u8ffd\u8e2a\u66f4\u65b0
\n if tracking:
\n # \u9010\u5e27\u66f4\u65b0\u8ffd\u8e2a\u5668，\u8fd4\u56de\u8ffd\u8e2a\u72b6\u6001\u548c\u76ee\u6807\u6846\u5750\u6807
\n success, box = tracker.update(frame)
\n if success:
\n x, y, w, h = [int(v) for v in box]
\n # \u7ed8\u5236\u8ffd\u8e2a\u6846（\u7eff\u8272，\u7ebf\u5bbd2）
\n cv2.rectangle(frame, (x, y), (x+w, y+h), (0,255,0), 2)<\/p>\n
# \u663e\u793a\u7ed3\u679c
\n cv2.imshow("CSRT Tracking", frame)<\/p>\n
cap.release()
\ncv2.destroyAllWindows() <\/p>\n
cv2.TrackerCSRT_create()：\u521b\u5efaCSRT\u8ffd\u8e2a\u5668\u5b9e\u4f8b，\u65e0\u9700\u989d\u5916\u914d\u7f6e，\u76f4\u63a5\u8c03\u7528\u5373\u53ef\u3002<\/p>\n<\/li>\n
cv2.selectROI()：\u5f39\u51fa\u7a97\u53e3\u8ba9\u7528\u6237\u624b\u52a8\u6846\u9009\u76ee\u6807\u533a\u57df，\u8fd4\u56de\u77e9\u5f62\u6846\u5750\u6807（x,y,w,h），`showCrosshair`\u53c2\u6570\u63a7\u5236\u662f\u5426\u663e\u793a\u5341\u5b57\u51c6\u661f\u3002<\/p>\n<\/li>\n
tracker.init()：\u521d\u59cb\u5316\u8ffd\u8e2a\u5668，\u7ed1\u5b9a\u521d\u59cb\u5e27\u548cROI，\u662f\u8ffd\u8e2a\u7684\u8d77\u70b9，\u9700\u5728\u9009\u62e9ROI\u540e\u8c03\u7528\u3002<\/p>\n<\/li>\n
tracker.update()：\u9010\u5e27\u66f4\u65b0\u8ffd\u8e2a\u72b6\u6001，\u8fd4\u56de\u4e24\u4e2a\u503c：`success`（\u5e03\u5c14\u503c，\u8ffd\u8e2a\u662f\u5426\u6210\u529f）\u548c`box`（\u76ee\u6807\u6846\u5750\u6807），\u5931\u8d25\u65f6`box`\u65e0\u6548\u3002<\/p>\n<\/li>\n
cv2.rectangle()：\u5728\u56fe\u50cf\u4e0a\u7ed8\u5236\u77e9\u5f62\u8ffd\u8e2a\u6846，\u53c2\u6570\u4f9d\u6b21\u4e3a：\u56fe\u50cf\u3001\u5de6\u4e0a\u89d2\u5750\u6807\u3001\u53f3\u4e0b\u89d2\u5750\u6807\u3001\u989c\u8272\u3001\u7ebf\u5bbd\u3002<\/p>\n<\/li>\n<\/ul>\n
\u6574\u5408\u7684\u5173\u952e\u662f\u89e3\u51b3\u4e24\u4e2a\u6838\u5fc3\u95ee\u9898：\u2460 \u8ffd\u8e2a\u7cbe\u5ea6：\u98ce\u683c\u8fc1\u79fb\u4f1a\u6539\u53d8\u56fe\u50cf\u50cf\u7d20，\u76f4\u63a5\u5728\u98ce\u683c\u5316\u56fe\u50cf\u4e0a\u8ffd\u8e2a\u4f1a\u5bfc\u81f4\u7cbe\u5ea6\u4e0b\u964d，\u56e0\u6b64\u91c7\u7528\u201c\u539f\u59cb\u5e27\u8ffd\u8e2a\u2192\u98ce\u683c\u5316\u5904\u7406\u2192\u8ffd\u8e2a\u6846\u7ed8\u5236\u5230\u98ce\u683c\u5316\u56fe\u50cf\u201d\u7684\u6d41\u7a0b；\u2461 \u683c\u5f0f\u517c\u5bb9：\u98ce\u683c\u5316\u7ed3\u679c\u9700\u8f6c\u4e3aOpenCV\u517c\u5bb9\u683c\u5f0f（3\u901a\u9053\u3001uint8\u3001\u8fde\u7eed\u5185\u5b58\u5e03\u5c40），\u907f\u514d\u7ed8\u56fe\u62a5\u9519\u3002<\/p>\n
import cv2
\nimport numpy as np<\/p>\n
# ====================== \u521d\u59cb\u5316\u914d\u7f6e ======================
\ncap = cv2.VideoCapture(0)
\nif not cap.isOpened():
\n print("\u65e0\u6cd5\u6253\u5f00\u6444\u50cf\u5934！")
\n exit()<\/p>\n
# \u521d\u59cb\u5316CSRT\u8ffd\u8e2a\u5668
\ntracker = cv2.TrackerCSRT_create()
\ntracking = False
\ntrack_box = None<\/p>\n
# \u52a0\u8f7d\u98ce\u683c\u8fc1\u79fb\u6a21\u578b
\ntry:
\n models = {
\n "left_top": cv2.dnn.readNet("model\/composition_vii.t7"),
\n "right_top": cv2.dnn.readNet("model\/candy.t7"),
\n "left_bottom": cv2.dnn.readNet("model\/udnie.t7"),
\n "right_bottom": cv2.dnn.readNet("model\/feathers.t7")
\n }
\n print("\u6a21\u578b\u52a0\u8f7d\u6210\u529f！")
\nexcept Exception as e:
\n print(f"\u6a21\u578b\u52a0\u8f7d\u5931\u8d25：{e}")
\n cap.release()
\n exit()<\/p>\n
# ====================== \u6838\u5fc3\u51fd\u6570 ======================
\ndef style_transfer(img, net):
\n h, w = img.shape[:2]
\n blob = cv2.dnn.blobFromImage(
\n img, 1.0, (w, h), (103.939, 116.779, 123.680),
\n swapRB=True, crop=False
\n )
\n net.setInput(blob)
\n preds = net.forward()
\n preds = preds.reshape(3, h, w)
\n preds = np.transpose(preds, (1, 2, 0))
\n preds = (preds – preds.min()) \/ (preds.max() – preds.min() + 1e-8)
\n preds = (preds * 255).astype(np.uint8)
\n if preds.shape[2] != 3:
\n preds = cv2.cvtColor(preds, cv2.COLOR_GRAY2BGR)
\n return preds<\/p>\n
# ====================== \u4e3b\u5faa\u73af ======================
\nwhile True:
\n ret, frame = cap.read()
\n if not ret:
\n break
\n frame_h, frame_w = frame.shape[:2]
\n if len(frame.shape) == 2:
\n frame = cv2.cvtColor(frame, cv2.COLOR_GRAY2BGR)<\/p>\n
# \u6309\u952e\u63a7\u5236
\n key = cv2.waitKey(1) & 0xFF
\n if key == ord('s'):
\n tracking = True
\n roi = cv2.selectROI("Tracking + Style Transfer", frame, showCrosshair=False)
\n tracker.init(frame, roi)
\n elif key == 27:
\n break<\/p>\n
# \u539f\u59cb\u5e27\u4e0a\u66f4\u65b0\u8ffd\u8e2a
\n if tracking:
\n success, box = tracker.update(frame)
\n track_box = [int(v) for v in box] if success else None<\/p>\n
# \u56db\u533a\u57df\u98ce\u683c\u8fc1\u79fb
\n h_half, w_half = frame_h\/\/2, frame_w\/\/2
\n tl = style_transfer(frame[:h_half, :w_half], models["left_top"])
\n tr = style_transfer(frame[:h_half, w_half:], models["right_top"])
\n bl = style_transfer(frame[h_half:, :w_half], models["left_bottom"])
\n br = style_transfer(frame[h_half:, w_half:], models["right_bottom"])<\/p>\n
# \u62fc\u63a5\u98ce\u683c\u5316\u56fe\u50cf（\u7edf\u4e00\u5c3a\u5bf8+\u8fde\u7eed\u5185\u5b58\u5e03\u5c40）
\n tl = cv2.resize(tl, (w_half, h_half))
\n tr = cv2.resize(tr, (w_half, h_half))
\n bl = cv2.resize(bl, (w_half, h_half))
\n br = cv2.resize(br, (w_half, h_half))
\n top_row = np.hstack((tl, tr))
\n bottom_row = np.hstack((bl, br))
\n styled_frame = np.vstack((top_row, bottom_row))
\n # \u5173\u952e：\u8f6c\u4e3a\u8fde\u7eed\u5185\u5b58\u5e03\u5c40，\u89e3\u51b3OpenCV\u7ed8\u56fe\u517c\u5bb9\u95ee\u9898
\n styled_frame = np.ascontiguousarray(styled_frame, dtype=np.uint8)<\/p>\n
# \u7ed8\u5236\u8ffd\u8e2a\u6846
\n if tracking and track_box is not None:
\n x, y, w, h = track_box
\n if 0 <= x < frame_w and 0 <= y < frame_h:
\n cv2.rectangle(styled_frame, (x, y), (x+w, y+h), (0,255,0), 2)<\/p>\n
cv2.imshow("Tracking + Style Transfer", styled_frame)<\/p>\n
cap.release()
\ncv2.destroyAllWindows() <\/p>\n
\u8ffd\u8e2a\u4e0e\u98ce\u683c\u5316\u987a\u5e8f：\u4f18\u5148\u5728\u539f\u59cb\u5e27\u4e0a\u6267\u884ctracker.update()，\u786e\u4fdd\u8ffd\u8e2a\u7cbe\u5ea6，\u518d\u5bf9\u539f\u59cb\u5e27\u505a\u98ce\u683c\u8fc1\u79fb，\u6700\u540e\u5c06\u8ffd\u8e2a\u6846\u7ed8\u5236\u5230\u98ce\u683c\u5316\u56fe\u50cf\u4e0a，\u907f\u514d\u98ce\u683c\u5316\u5e72\u6270\u8ffd\u8e2a\u3002<\/p>\n<\/li>\n
\u683c\u5f0f\u517c\u5bb9\u6027\u4fee\u590d：\u901a\u8fc7np.ascontiguousarray()\u5f3a\u5236\u98ce\u683c\u5316\u56fe\u50cf\u4e3a\u8fde\u7eed\u5185\u5b58\u5e03\u5c40，\u89e3\u51b3OpenCV cv2.rectangle() \u51fd\u6570\u62a5\u201cLayout incompatible\u201d\u9519\u8bef\u7684\u95ee\u9898\u3002<\/p>\n<\/li>\n
\u5c3a\u5bf8\u7edf\u4e00：\u7528cv2.resize() \u7edf\u4e00\u56db\u533a\u57df\u98ce\u683c\u5316\u7ed3\u679c\u7684\u5c3a\u5bf8，\u907f\u514d\u62fc\u63a5\u540e\u56fe\u50cf\u53d8\u5f62，\u786e\u4fdd\u8ffd\u8e2a\u6846\u5750\u6807\u6620\u5c04\u51c6\u786e\u3002<\/p>\n<\/li>\n
\u9c81\u68d2\u6027\u589e\u5f3a：\u589e\u52a0\u6444\u50cf\u5934\u6253\u5f00\u6821\u9a8c\u3001\u6a21\u578b\u52a0\u8f7d\u5f02\u5e38\u5904\u7406\u3001\u5750\u6807\u8d8a\u754c\u5224\u65ad，\u907f\u514d\u7a0b\u5e8f\u76f4\u63a5\u5d29\u6e83\u3002<\/p>\n<\/li>\n<\/ul>\n
\u6a21\u578b\u8def\u5f84：\u786e\u4fdd`model`\u6587\u4ef6\u5939\u4e0e\u4ee3\u7801\u6587\u4ef6\u540c\u7ea7，\u6a21\u578b\u6587\u4ef6\u540d\u4e0e\u4ee3\u7801\u4e2d\u4e00\u81f4，\u5426\u5219\u4f1a\u62a5\u6a21\u578b\u52a0\u8f7d\u5931\u8d25\u9519\u8bef\u3002<\/p>\n<\/li>\n
\u6027\u80fd\u4f18\u5316：\u82e5\u8fd0\u884c\u5361\u987f，\u53ef\u7f29\u5c0f\u6444\u50cf\u5934\u5206\u8fa8\u7387（\u5982`cap.set(cv2.CAP_PROP_FRAME_WIDTH, 640)`），\u51cf\u5c11\u98ce\u683c\u8fc1\u79fb\u7684\u8ba1\u7b97\u91cf\u3002<\/p>\n<\/li>\n
\u8ffd\u8e2a\u7a33\u5b9a\u6027：CSRT\u7cbe\u5ea6\u9ad8\u4f46\u901f\u5ea6\u7565\u6162，\u82e5\u9700\u5b9e\u65f6\u6027\u4f18\u5148，\u53ef\u66ff\u6362\u4e3acv2.TrackerKCF_create() \u8ffd\u8e2a\u5668\u3002<\/p>\n<\/li>\n
\u98ce\u683c\u8c03\u6574：\u66f4\u6362`.t7`\u6a21\u578b\u5373\u53ef\u5207\u6362\u98ce\u683c，\u53ef\u4eceOpenCV\u5b98\u65b9\u4ed3\u5e93\u6216\u7b2c\u4e09\u65b9\u8d44\u6e90\u83b7\u53d6\u66f4\u591a\u9884\u8bad\u7ec3\u6a21\u578b\u3002<\/p>\n<\/li>\n
\u672c\u6587\u901a\u8fc7\u6a21\u5757\u5316\u62c6\u89e3\u4e0e\u6574\u5408，\u5b9e\u73b0\u4e86\u4e24\u79cd\u6838\u5fc3\u89c6\u89c9\u4efb\u52a1\u7684\u878d\u5408，\u4ee3\u7801\u53ef\u76f4\u63a5\u843d\u5730\u8fd0\u884c\u3002\u5b9e\u9645\u9879\u76ee\u4e2d\u53ef\u57fa\u4e8e\u6b64\u6269\u5c55，\u5982\u589e\u52a0\u591a\u76ee\u6807\u8ffd\u8e2a\u3001\u98ce\u683c\u5207\u6362\u6309\u952e\u3001\u89c6\u9891\u6587\u4ef6\u8f93\u5165\u8f93\u51fa\u7b49\u529f\u80fd\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"
\u76ee\u5f55
\n\u4e00\u3001DNN\u98ce\u683c\u8fc1\u79fb：\u539f\u7406\u4e0e\u4ee3\u7801\u5b9e\u73b0
\n1. \u6838\u5fc3\u539f\u7406
\n2. \u5b9e\u6218\u4ee3\u7801\u5b9e\u73b0
\n3. \u5173\u952e\u51fd\u6570\u89e3\u6790
\n\u4e8c\u3001CSRT\u7269\u4f53\u8ffd\u8e2a：\u539f\u7406\u4e0e\u4ee3\u7801\u5b9e\u73b0
\n1. \u6838\u5fc3\u539f\u7406
\n2. \u5b9e\u6218\u4ee3\u7801\u5b9e\u73b0
\n3. \u5173\u952e\u51fd\u6570\u89e3\u6790
\n\u4e09\u3001\u4e24\u8005\u6574\u5408：\u5b9e\u65f6\u98ce\u683c\u8fc1\u79fb\u7269\u4f53\u8ffd\u8e2a
\n1. \u6574\u5408\u6838\u5fc3\u903b\u8f91
\n2. \u6574\u5408\u540e\u5b8c\u6574\u4ee3\u7801
\n3. \u6574\u5408\u5173\u952e\u4f18\u5316\u70b9
\n\u56db\u3001\u5b9e\u6218\u6ce8\u610f\u4e8b\u9879 \u5728\u8ba1\u7b97\u673a\u89c6\u89c9\u5b9e\u6218\u4e2d，\u5c06\u98ce\u683c\u8fc1\u79fb\u4e0e\u7269\u4f53\u8ffd\u8e2a\u7ed3\u5408\u53ef\u5b9e\u73b0\u66f4\u5177\u89c6\u89c9\u51b2\u51fb\u529b\u7684\u6548\u679c\u3002\u672c\u6587\u57fa\u4e8eOpen<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[6342,1460,50],"topic":[],"class_list":["post-66998","post","type-post","status-publish","format-standard","hentry","category-server","tag-dnn","tag-opencv","tag-50"],"yoast_head":"\n