Merge pull request #154 from liangxhao/update_readme

add readme for models list and benchmark

Author: SamitHuang

README.md

@@ -77,7 +77,9 @@ We will use **CRNN** model and **LMDB** dataset for demonstration, although othe
 
 #### Inference with MX Engine
 
-Please refer to [mx_infer](docs/cn/inference_cn.md).
+Please refer to [mx_infer tutorial](docs/cn/inference_tutorial_cn.md) for detailed inference tutorial.
+
+Please refer to [mx_infer results](docs/cn/inference_models_cn.md) for detailed performance of the supported inference models.
 
 #### Inference with Lite 
 

README_CN.md

@@ -69,7 +69,9 @@ MindOCR支持多种文本识别模型及数据集，在此我们使用**CRNN**
 
 #### 使用MX Engine推理
 
-请参考[mx_infer](docs/cn/inference_cn.md)
+教程请参考[mx_infer](docs/cn/inference_tutorial_cn.md)
+
+模型列表和Benchmark请参考 [mx_infer](docs/cn/inference_models_cn.md)
 
 #### 使用Lite推理 
 

docs/cn/inference_models_cn.md

@@ -0,0 +1,60 @@
+### MindOCR推理
+
+#### 支持模型列表
+
+##### 文本检测
+
+| 模型          | 链接                                                                                 | 来源        |
+|-------------|------------------------------------------------------------------------------------|-----------|
+| DBNet_Res18 | https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_infer.tar | PaddleOCR |
+| DBNet_MV3   | https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_infer.tar         | PaddleOCR |
+
+##### 文本方向分类
+
+| 模型  | 链接                                                                                 | 来源        |
+|-----|------------------------------------------------------------------------------------|-----------|
+| MV3 | https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar | PaddleOCR |
+
+##### 文本识别
+
+| 模型   | 链接                                                                                 | 来源        |
+|------|------------------------------------------------------------------------------------|-----------|
+| CRNN | https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_infer.tar | PaddleOCR |
+| SVTR | https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_infer.tar         | PaddleOCR |
+
+#### Benchmark
+
+1. 测试数据
+
+   数据集：ICDAR2019-LSVT
+
+   下载链接：https://rrc.cvc.uab.es/?ch=16
+
+   数据描述：30000张，街景图像，比如各种店铺招牌和地标等
+
+
+2. 测试环境
+
+   Device: Ascend310P
+
+   MindX: 3.0.0
+
+   CANN: 6.3
+
+   CPU: Intel(R) Xeon(R) Gold 6148, 2.40GHz, 2x20 physical cores
+
+3. 评估说明
+
+    - 性能：包括从图像输入到结果输出的完整阶段，设置mindocr推理命令--save_log_dir参数，保存日志中会记录性能数据
+    - 精度：参考评估脚本 mindocr/deploy/eval_utils/eval_script.py，输出结果包括文本检测的Precision、Recall和F-score，文本识别的Accuracy
+
+4. 精度和性能评估结果
+
+文本检测+方向分类+文本识别的端到端流水线评估结果如下：
+
+| 检测        | 分类 | 识别 | Precision | Recall | F-score | Accuracy | FPS   |
+| ----------- | ---- | ---- | --------- | ------ | ------- | -------- | ----- |
+| DBNet_Res18 | /    | CRNN | 69.42%    | 55.01% | 61.38%  | 47.12%   | 38.59 |
+| DBNet_Res18 | MV3  | CRNN | 69.42%    | 55.01% | 61.38%  | 46.85%   | 37.40 |
+| DBNet_MV3   | /    | SVTR | 67.01%    | 56.34% | 61.21%  | 46.77%   | 46.65 |
+| DBNet_MV3   | MV3  | SVTR | 67.01%    | 56.34% | 61.21%  | 46.45%   | 42.81 |