Merge pull request #19 from tink2123/add_words_img

add words_img

configs/rec/rec_icdar15_reader.yml

@@ -2,12 +2,12 @@ TrainReader:
   reader_function: ppocr.data.rec.dataset_traversal,SimpleReader
   num_workers: 8
   img_set_dir: ./train_data
-  label_file_path: ./train_data/rec_gt_train.txt
+  label_file_path: ./train_data/ic15_data/rec_gt_train.txt
   
 EvalReader:
   reader_function: ppocr.data.rec.dataset_traversal,SimpleReader
   img_set_dir: ./train_data
-  label_file_path: ./train_data/rec_gt_test.txt
+  label_file_path: ./train_data/ic15_data/rec_gt_test.txt
 
 TestReader:
   reader_function: ppocr.data.rec.dataset_traversal,SimpleReader

configs/rec/rec_icdar15_train.yml

 Global:
   algorithm: CRNN
   use_gpu: true
-  epoch_num: 3000
+  epoch_num: 1000
   log_smooth_window: 20
   print_batch_step: 10
   save_model_dir: ./output/rec_CRNN
@@ -14,7 +14,7 @@ Global:
   character_type: en
   loss_type: ctc
   reader_yml: ./configs/rec/rec_icdar15_reader.yml
-  pretrain_weights: ./pretrain_models/CRNN/best_accuracy 
+  pretrain_weights: ./pretrain_models/rec_mv3_none_bilstm_ctc/best_accuracy 
   checkpoints:
   save_inference_dir:
 Architecture:

doc/recognition.md

@@ -39,12 +39,13 @@ train_data/train_0002.jpg   用科技让复杂的世界更简单
 
 ```
 |-train_data
-    |- rec_gt_train.txt
-    |- train_imags
-        |- train_001.jpg
-        |- train_002.jpg
-        |- train_003.jpg
-        | ...
+    |-ic15_data
+        |- rec_gt_train.txt
+        |- train_imags
+            |- train_001.jpg
+            |- train_002.jpg
+            |- train_003.jpg
+            | ...
 ```
 
 - 评估集
@@ -53,12 +54,13 @@ train_data/train_0002.jpg   用科技让复杂的世界更简单
 
 ```
 |-train_data
-    |- rec_gt_eval.txt
-    |- eval_imags
-        |- eval_001.jpg
-        |- eval_002.jpg
-        |- eval_003.jpg
-        | ...
+    |-ic15_data
+        |- rec_gt_eval.txt
+        |- eval_imags
+            |- eval_001.jpg
+            |- eval_002.jpg
+            |- eval_003.jpg
+            | ...
 ```
 
 - 字典
@@ -120,16 +122,17 @@ PaddleOCR 提供了训练好的中文模型，可以[下载](todo: add)进行快
 默认预测图片存储在 `infer_img` 里，通过 `-o Global.checkpoints` 指定权重：
 
 ```
-python tools/infer_rec.py -c configs/rec/rec_chinese_lite_train.yml -o Global.checkpoints={path/to/weights}/best_accuracy
+python tools/infer_rec.py -c configs/rec/rec_chinese_lite_train.yml -o Global.checkpoints={path/to/weights}/best_accuracy TestReader.infer_img=doc/imgs_word/word_1.jpg
 ```
+预测图片：
+
+![](./imgs_words/word_1.jpg)
 
 得到输入图像的预测结果：
 
 ```
-infer_img: infer_img/328_4.jpg
-# 字符在字典中的索引
-[1863  921   55  155 1863 4209 3344  486  914 1863 4918]
-# 预测结果
-冷库专用冷冻液/载冷剂
+infer_img: doc/imgs_words/word_1.jpg
+     index: [2092  177  312 2503]
+     word : 韩国小馆
 ```
 

tools/infer_rec.py

@@ -37,7 +37,6 @@ set_paddle_flags(
 from paddle import fluid
 
 # from ppocr.utils.utility import load_config, merge_config
-from ppocr.data.reader_main import test_reader
 import program
 from paddle import fluid
 from ppocr.utils.utility import initial_logger