fix the co-ordication is minus bugs

lem89757 · May 27, 2020 · 41d8b55 · 41d8b55
1 parent 28e76fc
commit 41d8b55
Show file tree

Hide file tree

Showing 5 changed files with 41 additions and 21 deletions.
diff --git a/config.yml b/config.yml
@@ -1,6 +1,6 @@
 COMMON:
     WORKER : 3                  # 多少个进程同时生成图片
-    TEXT_CREATOR : corpus       # 随机生成:random, 语料生成:corpus
+    TEXT_CREATOR : random       # 随机生成:random, 语料生成:corpus
     GENERATOR : contour         # 保存轮廓的:contour, 只保存标签: text
 
 CORPUS:

diff --git a/syntext/augmentor.py b/syntext/augmentor.py
@@ -39,6 +39,11 @@ def augument(self, image, bbox_list):
                         alpha=(0, 3),  # 扭曲程度
                         sigma=(0.8, 1),  # 扭曲后的平滑程度
                         mode="nearest"),
+                ]),
+            ]),
+            # 模糊
+            iaa.Sometimes(0.5, [
+                iaa.OneOf([
                     iaa.GaussianBlur(sigma=(0, 0.7)),
                     iaa.AverageBlur(k=(1, 3)),
                     iaa.MedianBlur(k=(1, 3)),
@@ -94,6 +99,6 @@ def augument(self, image, bbox_list):
             polys = []
             for p in polygons_aug.polygons:
                 polys.append(p.coords)
-            polys = np.array(polys, np.int32).tolist()
+            polys = np.array(polys, np.int32).tolist() # (N,2)
 
         return image, polys
diff --git a/syntext/generator/generator.py b/syntext/generator/generator.py
@@ -36,7 +36,7 @@ def _caculate_position(self, text, font, x_offset, y_offset):
         char_bboxes = []
         for c in text:
             w, h = font.getsize(c)
-            if c == " ":
+            if c == " ": # 忽略空格，但是位置要空出来
                 x_offset += w
                 continue
             char_char_bbox = [
@@ -94,17 +94,30 @@ def _create_image(self, id, queue, num, dir):
                 image_path = os.path.join(dir, image_name)
                 image, text, bboxes = self._create_one_image()
                 image = self._pil2cv2(image)
+
+
                 image, bboxes = self.augmentor.augument(image, bboxes)
+
+                bboxes = self._revise_bboxes(bboxes)
+
                 label_data = self.build_label_data(text, bboxes)
                 cv2.imwrite(image_path, image)
-                debug_save_image(image_name, image, label_data)
+
+                debug_save_image(image_name, image, bboxes)
+
                 queue.put({'image': image_path, 'label': label_data})
             except Exception as e:
                 traceback.print_exc()
                 logger.error("[#%d-%d]样本生成发生错误，忽略此错误[%s]，继续...." % (id, i, str(e)))
 
         logger.info("生成进程[%d]生成完毕，合计[%d]张" % (id, num))
 
+    def _revise_bboxes(self, bboxes):
+        bboxes = np.array(bboxes)
+        minus_indices = bboxes < 0
+        bboxes[minus_indices] = 0
+        return bboxes.tolist()
+
     def _save_label(self, queue, total_num):
         counter = 0
         while True:
@@ -148,7 +161,7 @@ def _create_one_image(self):
         logger.debug("贴文字到背景的位置：(%d,%d)", x, y)
         background.paste(image, (x, y), image)
 
-        bboxes = self._caculate_position(text, font, x, y)
+        bboxes = self._caculate_position(text, font, x, y) # 计算每个字的bbox
 
         return background, text, bboxes
 

diff --git a/syntext/generator/generator_builder.py b/syntext/generator/generator_builder.py
@@ -3,24 +3,35 @@
 from syntext.text.corpus_creator import CorpusTextCreator
 from syntext.text.random_creator import RandomTextCreator
 from syntext.augmentor import Augumentor
+import logging
 
+logger = logging.getLogger(__name__)
 
 class GeneratorBuilder():
 
     def build(self, config, charset, fonts, backgrounds):
+        logger.info("创建生成器：")
+        logger.info("----------------------------------------")
+
         # 图像增强
         augmentor = Augumentor(config)
+        logger.info("\t创建增强器\t：使用imgaug做图像增强")
 
         # 文本创建器
         if config.COMMON['TEXT_CREATOR'] == "random":
+            logger.info("\t创建文本生成器\t：随机产生文本")
             text_creator = RandomTextCreator(config, charset)
         if config.COMMON['TEXT_CREATOR'] == "corpus":
             text_creator = CorpusTextCreator(config)
+            logger.info("\t创建文本生成器\t：从语料中抽取文本")
 
         # 标签保存器
         if config.COMMON['GENERATOR'] == "contour":
             generator = ContourGenerator(config, charset, fonts, backgrounds, text_creator, augmentor)
+            logger.info("\t创建样本保存器\t：保存每个字的坐标")
         if config.COMMON['GENERATOR'] == "text":
             generator = TextOnlyGenerator(config, charset, fonts, backgrounds, text_creator, augmentor)
+            logger.info("\t创建样本保存器\t：仅保存字符串")
 
+        logger.info("----------------------------------------")
         return generator
diff --git a/syntext/utils/utils.py b/syntext/utils/utils.py
@@ -46,7 +46,7 @@ def debug(*args):
 
 
 
-def debug_save_image(name, image, label):
+def debug_save_image(name, image, bboxes): # bboxes[N,M,2]
     if not DEBUG: return
 
     image = image.copy()
@@ -55,21 +55,12 @@ def debug_save_image(name, image, label):
         os.makedirs(debug_dir)
     debug_image_path = os.path.join(debug_dir, name)
 
-    # {
-    #   "label":"你好世界！",
-    #   "pos":[
-    #       {
-    #           "bbox": [[x1,y1],[x1,y1],[x1,y1],[x1,y1]],
-    #           "word": "你"
-    #       },
-    #       ....
-    #   ]
-    # }
-    if type(label)==dict and "pos" in label:
-        for pos in label['pos']:
-            bboxes = np.array(pos['bbox'])
-            cv2.polylines(image, [bboxes], isClosed=True, color=(0,0,255))
-        cv2.imwrite(debug_image_path,image)
+    for one_word_bboxes in bboxes:
+        # print(one_word_bboxes)
+        one_word_bboxes = np.array(one_word_bboxes)
+        cv2.polylines(image, [one_word_bboxes], isClosed=True, color=(0, 0, 255))
+
+    cv2.imwrite(debug_image_path,image)
 
 
 def dynamic_load(module_name, parent_class):