paddlecor检测可视化标注

推理缩放尺寸：

训练数据预处理增强

EastRandomCropData代码：

paddlecor检测可视化标注

import os

import cv2

dir_path=r'E:\project\icdar_c4_train_tmp/'

labelpath=dir_path+'/Label.txt'
labeldict = {}
with open(labelpath, 'r', encoding='utf-8') as f:
    data = f.readlines()
    for each in data:
        file, label = each.split('\t')
        if label:
            label = label.replace('false', 'False')
            label = label.replace('true', 'True')
            labeldict[file] = eval(label)
        else:
            labeldict[file] = []

for k,vs in labeldict.items():

    img=cv2.imread(dir_path+k)
    for v in vs:
        if "transcription" in v:
            for index, point in  enumerate(v['points']):

                color=(255, 0, 0)
                if index==1:
                    color = (255, 255, 0)
                if index == 2:
                    color = (255, 0, 255)
                if index == 3:
                    color = (0, 0, 255)
                cv2.circle(img, (point[0], point[1]), 1, color, 2)
            print(k,v['points'])

            cv2.imshow("asdf",img)
            cv2.waitKey()

推理缩放尺寸：

检测用的：resize_image_type1

DetResizeForTest参数： h，w。

transforms:
#      - DecodeImage: # load image
#          img_mode: BGR
#          channel_first: False
      - DetLabelEncode: # Class handling label
      - DetResizeForTest:
           image_shape: [128, 352]

缩放代码：

operators.py中：

class DetResizeForTest(object):
    def __init__(self, **kwargs):
        super(DetResizeForTest, self).__init__()
        self.resize_type = 0
        if 'image_shape' in kwargs:
            self.image_shape = kwargs['image_shape']
            self.resize_type = 1
        elif 'limit_side_len' in kwargs:
            self.limit_side_len = kwargs['limit_side_len']
            self.limit_type = kwargs.get('limit_type', 'min')
        elif 'resize_long' in kwargs:
            self.resize_type = 2
            self.resize_long = kwargs.get('resize_long', 960)
        else:
            self.limit_side_len = 736
            self.limit_type = 'min'

    def __call__(self, data):
        img = data['image']
        src_h, src_w, _ = img.shape

        if self.resize_type == 0:
            # img, shape = self.resize_image_type0(img)
            img, [ratio_h, ratio_w] = self.resize_image_type0(img)
        elif self.resize_type == 2:
            img, [ratio_h, ratio_w] = self.resize_image_type2(img)
        else:
            # img, shape = self.resize_image_type1(img)
            img, [ratio_h, ratio_w] = self.resize_image_type1(img)
        data['image'] = img
        data['shape'] = np.array([src_h, src_w, ratio_h, ratio_w])
        return data

    def resize_image_type1(self, img):
        resize_h, resize_w = self.image_shape
        ori_h, ori_w = img.shape[:2]  # (h, w, c)
        ratio_h = float(resize_h) / ori_h
        ratio_w = float(resize_w) / ori_w
        img = cv2.resize(img, (int(resize_w), int(resize_h)))
        # return img, np.array([ori_h, ori_w])
        return img, [ratio_h, ratio_w]

按比例缩放版：resize_image_type1

def resize_image_type1(self, img):
        # resize_h, resize_w = self.image_shape
        # ori_h, ori_w = img.shape[:2]  # (h, w, c)
        # ratio_h = float(resize_h) / ori_h
        # ratio_w = float(resize_w) / ori_w
        # img = cv2.resize(img, (int(resize_w), int(resize_h)))

        t_h, t_w = img.shape[:2]

        to_w = 352
        to_h = 128

        img_b = np.zeros((to_h, to_w, 3), dtype=np.uint8)
        if t_h / t_w > to_h / to_w:
            x_scale = to_h / img.shape[0]
            img = cv2.resize(img, None, fx=x_scale, fy=x_scale, interpolation=cv2.INTER_AREA)
            t_h, t_w = img.shape[:2]
            img_b[:, (to_w - t_w) // 2: (t_w + to_w) // 2, :] = img
        else:
            x_scale = to_w / img.shape[1]
            img = cv2.resize(img, None, fx=x_scale, fy=x_scale, interpolation=cv2.INTER_AREA)
            t_h, t_w = img.shape[:2]
            img_b[(to_h - t_h) // 2:t_h + (to_h - t_h) // 2, :, :] = img
        cv2.imshow('resize',img_b)
        # return img, np.array([ori_h, ori_w])
        return img_b, [x_scale, x_scale]

训练数据预处理增强

配置文件ch_det_mv3_db_v2.0.yml参数：

EastRandomCropData 参数：h w

因为NormalizeImage的参数为hwc

去掉了flip增强，减小了Resize比例，从[0.5-3]改为了[0.8,1.5]

修改了EastRandomCropData 缩放宽高

transforms:
#      - DecodeImage: # load image
#          img_mode: BGR
#          channel_first: False
      - DetLabelEncode: # Class handling label
      - IaaAugment:
          augmenter_args:
#            - { 'type': Fliplr, 'args': { 'p': 0.5 } }
            - { 'type': Affine, 'args': { 'rotate': [-5, 5] } }
            - { 'type': Resize, 'args': { 'size': [0.8, 1.5] } }
      - EastRandomCropData:
          size:  [128, 352] # w h
          max_tries: 50
          keep_ratio: true
      - MakeBorderMap:
          shrink_ratio: 0.4
          thresh_min: 0.3
          thresh_max: 0.7
      - MakeShrinkMap:
          shrink_ratio: 0.4
          min_text_size: 12
      - NormalizeImage:
          scale: 1./255.
          mean: [0.485, 0.456, 0.406]
          std: [0.229, 0.224, 0.225]
          order: 'hwc'
      - ToCHWImage:
      - KeepKeys:
          keep_keys: ['image', 'threshold_map', 'threshold_mask', 'shrink_map', 'shrink_mask'] # the order of the dataloader list

EastRandomCropData代码：

这里面的size 顺序 w h，

class EastRandomCropData(object):
    def __init__(self, size=(640, 640), max_tries=10, min_crop_side_ratio=0.3, keep_ratio=True, **kwargs):
        self.size = size
        self.size[0],self.size[1]=self.size[1],self.size[0]
        self.max_tries = max_tries
        self.min_crop_side_ratio = min_crop_side_ratio
        self.keep_ratio = keep_ratio

    def __call__(self, data):
        img = data['image']
        text_polys = data['polys']
        ignore_tags = data['ignore_tags']
        texts = data['texts']
        all_care_polys = [text_polys[i] for i, tag in enumerate(ignore_tags) if not tag]
        # 计算crop区域
        crop_x, crop_y, crop_w, crop_h = crop_area(img, all_care_polys, self.min_crop_side_ratio, self.max_tries)
        # crop 图片 保持比例填充
        scale_w = self.size[0] / crop_w
        scale_h = self.size[1] / crop_h
        scale = min(scale_w, scale_h)
        h = int(crop_h * scale)
        w = int(crop_w * scale)
        if self.keep_ratio:
            padimg = np.zeros((self.size[1], self.size[0], img.shape[2]), img.dtype)
            padimg[:h, :w] = cv2.resize(img[crop_y:crop_y + crop_h, crop_x:crop_x + crop_w], (w, h))
            # img_a=cv2.resize(img[crop_y:crop_y + crop_h, crop_x:crop_x + crop_w], (w, h))
            # print(img_a.shape)
            # cv2.imshow("crop_area",img_a)
            # cv2.waitKey()
            img = padimg
        else:
            img = cv2.resize(img[crop_y:crop_y + crop_h, crop_x:crop_x + crop_w], tuple(self.size))
        # crop 文本框
        text_polys_crop = []
        ignore_tags_crop = []
        texts_crop = []
        for poly, text, tag in zip(text_polys, texts, ignore_tags):
            poly = ((poly - (crop_x, crop_y)) * scale).tolist()
            if not is_poly_outside_rect(poly, 0, 0, w, h):
                text_polys_crop.append(poly)
                ignore_tags_crop.append(tag)
                texts_crop.append(text)
        data['image'] = img
        data['polys'] = np.array(text_polys_crop)
        data['ignore_tags'] = ignore_tags_crop
        data['texts'] = texts_crop
        return data