diff --git a/.github/workflows/wired_table_rec.yml b/.github/workflows/wired_table_rec.yml
index 1c5f5a2..fc65e1b 100644
--- a/.github/workflows/wired_table_rec.yml
+++ b/.github/workflows/wired_table_rec.yml
@@ -33,10 +33,6 @@ jobs:
           unzip wired_table_rec_models.zip
           mv wired_table_rec_models/*.onnx wired_table_rec/models/
 
-          wget https://github.com/RapidAI/TableStructureRec/releases/download/v0.0.0/wired_table_rec_modelsV2.zip
-          unzip wired_table_rec_modelsV2.zip
-          mv cycle_center_net_v2.onnx wired_table_rec/models/
-
           pytest tests/test_wired_table_rec.py
 
   GenerateWHL_PushPyPi:
diff --git a/README.md b/README.md
index 9deafa6..b0ff3fb 100644
--- a/README.md
+++ b/README.md
@@ -16,7 +16,9 @@
 - **2024.9.26**
   - 修正RapidTable默认英文模型导致的测评结果错误。
   - 补充测评数据集，补充开源社区更多模型的测评结果
-
+- **2024.9.30**
+  - 优化表格类型判断，增加在线演示
+    
 ### 简介
 💖该仓库是用来对文档中表格做结构化识别的推理库，包括来自paddle的表格识别模型，
 阿里读光有线和无线表格识别模型，llaipython(微信)贡献的有线表格模型，网易Qanything内置表格分类模型等。
@@ -29,6 +31,9 @@
 
 🛡️ **稳**: 不依赖任何第三方训练框架，只依赖必要基础库，避免包冲突
 
+### 在线演示
+[modelscope魔塔](https://www.modelscope.cn/studios/jockerK/TableRec)
+[huggingface](https://huggingface.co/spaces/Joker1212/TableDetAndRec)
 ### 效果展示
 
 <div align="center">
diff --git a/table_cls/main.py b/table_cls/main.py
index 9209e2a..9ad7820 100644
--- a/table_cls/main.py
+++ b/table_cls/main.py
@@ -1,6 +1,7 @@
 import time
 from pathlib import Path
 
+import cv2
 import numpy as np
 import onnxruntime
 from PIL import Image
@@ -39,8 +40,10 @@ def _preprocess(self, image):
     def __call__(self, content: InputType):
         ss = time.perf_counter()
         img = self.load_img(content)
-        img = self._preprocess(img)
-        output = self.table_cls.run(None, {"input": img})
+        gray_img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
+        gray_img = np.stack((gray_img,) * 3, axis=-1)
+        gray_img = self._preprocess(gray_img)
+        output = self.table_cls.run(None, {"input": gray_img})
         predict = np.exp(output[0] - np.max(output[0], axis=1, keepdims=True))
         predict /= np.sum(predict, axis=1, keepdims=True)
         predict_cla = np.argmax(predict, axis=1)[0]