def detect_box(image, scale=600, maxScale=900):
H, W = image.shape[:2]
image, rate = resize_img(image, scale, maxScale=maxScale)
h, w = image.shape[:2]
if GPU:
im = array_to_image(image)
res = predict_image(textNet, im)
scale = 16
iw = int(np.ceil(im.w / scale))
ih = int(np.ceil(im.h / scale))
h, w = image.shape[:2]
out = [res[i] for i in range(40 * ih * iw)]
out = np.array(out).reshape((1, 40, ih, iw))
else:
inputBlob = cv2.dnn.blobFromImage(image,
scalefactor=1.0,
size=(w, h),
swapRB=False,
crop=False)
outputName = textNet.getUnconnectedOutLayersNames()
textNet.setInput(inputBlob)
out = textNet.forward(outputName)[0]
clsOut = reshape(out[:, :20, ...])
boxOut = reshape(out[:, 20:, ...])
boxes = get_origin_box((w, h), anchors, boxOut[0])
scores = soft_max(clsOut[0])
boxes[:, 0:4][boxes[:, 0:4] < 0] = 0
boxes[:, 0][boxes[:, 0] >= w] = w - 1
boxes[:, 1][boxes[:, 1] >= h] = h - 1
boxes[:, 2][boxes[:, 2] >= w] = w - 1
boxes[:, 3][boxes[:, 3] >= h] = h - 1
# print (boxes)
return scores, boxes, rate, w, h
def predict_darknet(image):
scale = image.size[1] * 1.0 / 32
w = image.size[0] / scale
w = int(w)
image = image.resize((w, 32), Image.BILINEAR)
image = (np.array(image.convert('L')) / 255.0 - 0.5) / 0.5
h, w = image.shape
if w < 8:
return ''
tmp = np.zeros((h, w, 1))
tmp[:, :, 0] = image
im = array_to_image(image)
res = predict_image(ocrNet, im)
outW = int(np.ceil(w / 4) - 3)
nchars = len(charactersPred)
out = [res[i] for i in range(outW * nchars)]
out = np.array(out).reshape((nchars, outW))
out = out.transpose((1, 0))
return decode(out)
def predict_darknet(image):
scale = image.size[1] * 1.0 / 32
w = image.size[0] / scale
w = int(w)
image = image.resize((w, 32), Image.BILINEAR)
image = (np.array(image.convert('L')) / 255.0 - 0.5) / 0.5
h, w = image.shape
if w < 8:
return {'chars': [], 'text': '', 'prob': 0}
tmp = np.zeros((h, w, 1))
tmp[:, :, 0] = image
im = array_to_image(image)
res = predict_image(ocrNet, im)
outW = int(np.ceil(w / 4) - 3)
nchars = len(charactersPred)
out = np.zeros(outW * nchars, dtype=res._type_)
memmove(out.ctypes.data, res, out.nbytes)
out = out.reshape((nchars, outW))
out = out.transpose((1, 0))
out = softmax(out)
return decode(out)