def get_det_boxes(image, display=True):
h, w = image.shape[:2]
if w > max_width:
image = resize(image, width=max_width)
elif w < min_width:
image = resize(image, width=min_width)
image_r = image.copy()
image_c = image.copy()
h, w = image.shape[:2]
image = image.astype(np.float32) - config.IMAGE_MEAN
image = torch.from_numpy(image.transpose(2, 0, 1)).unsqueeze(0).float()
with torch.no_grad():
image = image.to(device)
cls, regr = model(image)
cls_prob = F.softmax(cls, dim=-1).cpu().numpy()
regr = regr.cpu().numpy()
anchor = gen_anchor((int(h / 16), int(w / 16)), 16)
bbox = bbox_transfor_inv(anchor, regr)
bbox = clip_box(bbox, [h, w])
# print(bbox.shape)
fg = np.where(cls_prob[0, :, 1] > prob_thresh)[0]
# print(np.max(cls_prob[0, :, 1]))
select_anchor = bbox[fg, :]
select_score = cls_prob[0, fg, 1]
select_anchor = select_anchor.astype(np.int32)
# print(select_anchor.shape)
keep_index = filter_bbox(select_anchor, 16)
# nms
select_anchor = select_anchor[keep_index]
select_score = select_score[keep_index]
select_score = np.reshape(select_score, (select_score.shape[0], 1))
nmsbox = np.hstack((select_anchor, select_score))
keep = nms(nmsbox, 0.3)
# print(keep)
select_anchor = select_anchor[keep]
select_score = select_score[keep]
# text line-
textConn = TextProposalConnectorOriented()
text = textConn.get_text_lines(select_anchor, select_score, [h, w])
# print(text)
if display:
for i in text:
s = str(round(i[-1] * 100, 2)) + '%'
i = [int(j) for j in i]
cv2.line(image_c, (i[0], i[1]), (i[2], i[3]), (0, 0, 255), 2)
cv2.line(image_c, (i[0], i[1]), (i[4], i[5]), (0, 0, 255), 2)
cv2.line(image_c, (i[6], i[7]), (i[2], i[3]), (0, 0, 255), 2)
cv2.line(image_c, (i[4], i[5]), (i[6], i[7]), (0, 0, 255), 2)
cv2.putText(image_c, s, (i[0] + 13, i[1] + 13),
cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 0, 0), 2,
cv2.LINE_AA)
return text, image_c, image_r
def test(img_path):
image = cv2.imread(img_path)
# ipdb.set_trace()
image_c = image.copy()
# h1,w1,c=image_c.shape
# oddnumber=w1/width
# image = resize(image, width=width)
h, w = image.shape[:2]
image = image.astype(np.float32) - config.IMAGE_MEAN
image = torch.from_numpy(image.transpose(2, 0, 1)).unsqueeze(0).float()
with torch.no_grad():
image = image.to(device)
cls, regr = model(image)
cls_prob = F.softmax(cls, dim=-1).cpu().numpy()
regr = regr.cpu().numpy()
anchor = gen_anchor((math.ceil(h / 16), math.ceil(w / 16)), 16)
bbox = bbox_transfor_inv(anchor, regr)
bbox = clip_box(bbox, [h, w])
fg = np.where(cls_prob[0, :, 1] > prob_thresh)[0]
select_anchor = bbox[fg, :]
select_score = cls_prob[0, fg, 1]
select_anchor = select_anchor.astype(np.int32)
keep_index = filter_bbox(select_anchor, 16)
# nsm
select_anchor = select_anchor[keep_index]
select_score = select_score[keep_index]
select_score = np.reshape(select_score, (select_score.shape[0], 1))
nmsbox = np.hstack((select_anchor, select_score))
keep = nms(nmsbox, 0.3)
select_anchor = select_anchor[keep]
select_score = select_score[keep]
# text line-
textConn = TextProposalConnectorOriented()
text = textConn.get_text_lines(select_anchor, select_score, [h, w])
# print(text)
alltext = []
for i in text:
if i[-1] >= 0.9:
s = str(round(i[-1] * 100, 2)) + '%'
i = [int(j) for j in i]
cv2.line(image_c, (i[0], i[1]), (i[2], i[3]), (0, 0, 255), 2)
cv2.line(image_c, (i[0], i[1]), (i[4], i[5]), (0, 0, 255), 2)
cv2.line(image_c, (i[6], i[7]), (i[2], i[3]), (0, 0, 255), 2)
cv2.line(image_c, (i[4], i[5]), (i[6], i[7]), (0, 0, 255), 2)
cv2.putText(image_c, s, (i[0] + 13, i[1] + 13),
cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 0, 0), 1,
cv2.LINE_AA)
imagename = cutstr(img_path)
savepicpath = os.path.join(EPOCH_IMAGE, imagename + '.png')
cv2.imwrite(savepicpath, image_c)
save_path = os.path.join(EPOCH_TXT, imagename + '.txt')
file = open(save_path, 'w')
file.write(str(alltext))
file.close()
def test(img_path):
image = cv2.imread(img_path)
isize = image.shape
# ipdb.set_trace()
image_c = image.copy()
# h1,w1,c=image_c.shape
# oddnumber=w1/width
# image = resize(image, width=width)
h, w = image.shape[:2]
image = image.astype(np.float32) - config.IMAGE_MEAN
image = torch.from_numpy(image.transpose(2, 0, 1)).unsqueeze(0).float()
with torch.no_grad():
image = image.to(device)
cls, regr = model(image)
cls_prob = F.softmax(cls, dim=-1).cpu().numpy()
regr = regr.cpu().numpy()
anchor = gen_anchor((math.ceil(h / 16), math.ceil(w / 16)), 16)
bbox = bbox_transfor_inv(anchor, regr)
bbox = clip_box(bbox, [h, w])
fg = np.where(cls_prob[0, :, 1] > prob_thresh)[0]
boxes = bbox[fg, :] # 可用的框格
scores = cls_prob[0, fg, 1]
select_anchor = boxes.astype(np.float32)
keep_index = filter_bbox(select_anchor, 16)
# nsm
boxes = select_anchor[keep_index]
scores = scores[keep_index]
NMS_THRESH = 0.3
dets = np.hstack((boxes, scores[:, np.newaxis])).astype(np.float32)
keep = nms(dets, NMS_THRESH)
dets = dets[keep, :]
keep = np.where(dets[:, 4] >= 0.7)[0]
dets = dets[keep, :]
line = connect_proposal(dets[:, 0:4], dets[:, 4], isize)
save_results(img_path, line, thresh=0.9)
def inference(self, image):
image_sz = resize(image, height=ctpn_params.IMAGE_HEIGHT)
# 宽高缩放比例(等比例缩放)
rescale_fac = image.shape[0] / image_sz.shape[0]
h, w = image_sz.shape[:2]
# 减均值
image_sz = image_sz.astype(np.float32) - ctpn_params.IMAGE_MEAN
image_sz = torch.from_numpy(image_sz.transpose(
2, 0, 1)).unsqueeze(0).float()
if self.use_gpu:
image_sz = image_sz.cuda()
cls, regr = self.model(image_sz)
cls_prob = F.softmax(cls, dim=-1).cpu().numpy()
regr = regr.cpu().numpy()
anchor = gen_anchor((int(h / 16), int(w / 16)), 16)
bbox = bbox_transfor_inv(anchor, regr)
bbox = clip_box(bbox, [h, w])
fg = np.where(cls_prob[0, :, 1] > self.prob_thresh)[0]
select_anchor = bbox[fg, :]
select_score = cls_prob[0, fg, 1]
select_anchor = select_anchor.astype(np.int32)
keep_index = filter_bbox(select_anchor, 16)
# nms
select_anchor = select_anchor[keep_index]
select_score = select_score[keep_index]
select_score = np.reshape(select_score, (select_score.shape[0], 1))
nmsbox = np.hstack((select_anchor, select_score))
keep = nms(nmsbox, 0.3)
select_anchor = select_anchor[keep]
select_score = select_score[keep]
# text line-
textConn = TextProposalConnectorOriented()
text = textConn.get_text_lines(select_anchor, select_score, [h, w])
text = [np.hstack((res[:8] * rescale_fac, res[8])) for res in text]
return text
image = cv2.imread(filenames[i])
image = resize(image, width=width)
image_c = image.copy()
h, w = image.shape[:2]
image = image.astype(np.float32) - config.IMAGE_MEAN
image = torch.from_numpy(image.transpose(2, 0, 1)).unsqueeze(0).float()
with torch.no_grad():
image = image.to(device)
cls, regr = model(image)
cls_prob = F.softmax(cls, dim=-1).cpu().numpy()
regr = regr.cpu().numpy()
anchor = gen_anchor((int(h / 16), int(w / 16)), 16)
bbox = bbox_transfor_inv(anchor, regr)
bbox = clip_box(bbox, [h, w])
# prob_thresh = 0.3
# print(cls_prob)
# print(cls_prob[0, :, 1])
# print(cls_prob.shape)
fg = np.where(cls_prob[0, :, 1] > prob_thresh)[0]
#print("fg:", fg)
select_anchor = bbox[fg, :]
print("select_anchor:")