Toybrick

标题: mtcnn推理图片尺寸 [打印本页]

作者: liuwenhua 时间: 2019-12-23 15:38
标题: mtcnn推理图片尺寸
capture = cv2.VideoCapture(0)
while (True):
      tic()
      ret, img = capture.read()
      if ret == True:
         #img = cv2.imread('./test4.jpg')
         print(img.shape)
         img = cv2.resize(img,(450,344))
         #img = cv2.resize(img,(640,480))
         img_matlab = img.copy()
         img_matlab = cv2.cvtColor(img_matlab, cv2.COLOR_BGR2RGB)
         toc()
         print("capture--------------------")
         #img_matlab = img.copy()
         #tmp = img_matlab[:,:,2].copy()
         #img_matlab[:,:,2] = img_matlab[:,:,0]
         #img_matlab[:,:,0] = tmp
         tic()
         # check rgb position
         boundingboxes, points = detect_face(img_matlab, minsize, pnet_rknn_list, rnet_rknn, onet_rknn, threshold, False, factor)
         toc()
         tic()
         img = drawBoxes(img, boundingboxes)
推理时图片尺寸必须为(450,344)，可以改为自己想要的尺寸？如果可以改的话，应该怎么进行修改，谁遇到类似问题，求指导

作者: jefferyzhang 时间: 2019-12-24 08:22
改shape需要改模型，rknn模型的输入、输出、batch都必须是固定的静态图

作者: liuwenhua 时间: 2019-12-24 08:52
问题已解决，谢谢

作者: zengwubin 时间: 2020-3-20 11:18

liuwenhua 发表于 2019-12-24 08:52
问题已解决，谢谢

请问，能说下你是怎么做的吗？

作者: liuwenhua 时间: 2020-3-24 15:12

zengwubin 发表于 2020-3-20 11:18
请问，能说下你是怎么做的吗？

PNET_PYRAMID= np.array([[1152,648],[817,460],[580,326],[411,231],[292,164],[207,117],[147,83],[104,59],[74,42],[53,30],
[37,21],[27,15],[19,11]])
PNET_PYRAMID_ARR= np.array([[[1, 2, 571, 319],[1, 4, 571, 319]],[[1, 2, 404, 225],[1, 4, 404, 225]],
[[1,2,285,158],[1,4,285,158]],[[1, 2,201,111],[1, 4, 201,111]],[[1, 2, 141, 77],[1,4,141,77]],[[1,2,99,54],[1, 4,99,54]],
[[1,2,69,37],[1,4,69,37]],[[1,2,47,25],[1,4,47,25]],[[1,2,32,16],[1,4,32,16]],[[1,2,20,11],[1,4,20,11]],
[[1,2,14,6],[1,4,14,6]],[[1, 2, 9, 3],[1, 4, 9, 3]],[[1, 2, 5, 1],[1, 4, 5, 1]]])

作者: liuwenhua 时间: 2020-3-24 15:13

liuwenhua 发表于 2020-3-24 15:12
PNET_PYRAMID= np.array([[1152,648],[817,460],[580,326],[411,231],[292,164],[207,117],[147,83],[104 ...

根据自己输入图片的尺寸，生成自己的p网络

欢迎光临 Toybrick (https://t.rock-chips.com/)