111

README.md

@@ -1,14 +1,24 @@
 ---
 license: apache-2.0
 ---
-# 基于FormalGeo7K的结构识别模型
+# Diagram Formalizer
+Model Structure: 
 
-## 快速开始
-在运行脚本之前，首先安装如下必要的依赖。
+<img src="sample/diagram_formalizer.png" alt="Alt text" width="20%" height="auto">
+
+Diagram Encoder: [siglip-so400m-patch14-384](https://huggingface.co/google/siglip-so400m-patch14-384)
+
+Lightweight LLM: [Qwen2-0.5B-Instruct](https://huggingface.co/Qwen/Qwen2-0.5B-Instruct)
+
+
+
+## Quick Start
+Before running the script, install the following necessary dependencies.
 
 ```shell
 pip install torch transformers==4.40.0 accelerate pillow sentencepiece
 ```
+You can use the following script to predict the ConsCDL and ImgCDL for geometric diagram.
 
 ```python
 import torch
@@ -24,12 +34,12 @@ torch.set_default_device(device)
 
 # create model
 model = AutoModelForCausalLM.from_pretrained(
-    'NaughtyDog97/GeoFormalizer',
+    'NaughtyDog97/DiagramFormalizer',
     torch_dtype=torch.float16, # float32 for cpu
     device_map='auto',
     trust_remote_code=True)
 tokenizer = AutoTokenizer.from_pretrained(
-    'NaughtyDog97/GeoFormalizer',
+    'NaughtyDog97/DiagramFormalizer',
     use_fast=True,
     padding_side="right",
     trust_remote_code=True)
@@ -89,41 +99,28 @@ print(respones)
 
 ```
 
-我们的模型支持的识别方式有如下几种:
-- 自然语言描述: 
+Our model supports the following recognition instrutions:
+- Natural Language Description: 
     - Describe what you see in the figure.
     - Tell me what you observe in the image.
-    - 使用自然语言描述这幅图像。
-- 只预测construction_cdl
+- Predicting ConsCDL only
     - Based on the image, predict the construction_cdl.
-    - 根据图像识别出construction_cdl。
     - Based on the image, predict the construction_cdl and calibrate it.
-    - 根据图像识别出construction_cdl并进行矫正。
     - Based on the image, first describe what you see in the figure, then predict the construction_cdl.
-    - 根据图像，首先描述图像，之后识别出construction_cdl。
     - Based on the image, first describe what you see in the figure, then predict the construction_cdl and calibrate it.
-    - 根据图像，首先描述图像，之后识别出construction_cdl并进行矫正。
-- 只预测image_cdl
+- Predicting ImgCDL only:
     - Based on the image, predict the image_cdl.
-    - 根据图像识别出image_cdl。
     - Based on the image, predict the image_cdl and calibrate it.
-    - 根据图像识别出image_cdl并进行矫正。
     - Based on the image, first describe what you see in the figure, then predict the image_cdl.
-    - 根据图像，首先描述图像，之后识别出image_cdl。
     - Based on the image, first describe what you see in the figure, then predict the image_cdl and calibrate it.
-    - 根据图像，首先描述图像，之后识别出image_cdl并进行矫正。
-- 同时预测construction_cdl和image_cdl
+- Predicting construction_cdl and image_cdl simultaneously:
     - Based on the image, predict the construction_cdl and image_cdl.
-    - 根据图像识别出construction_cdl和image_cdl。
     - Based on the image, first predict the construction_cdl and image_cdl and calibrate it.
-    - 根据图像识别出construction_cdl和image_cdl并进行矫正。
     - Based on the image, first describe what you see in the figure, then predict the construction_cdl and image_cdl.
-    - 根据图像，首先描述图像，之后识别出construction_cdl和image_cdl。
     - Based on the image, first describe what you see in the figure, then predict the construction_cdl and image_cdl and calibrate it.
-    - 根据图像，首先描述图像，之后识别出construction_cdl和image_cdl并矫正。
 
 
-## Performance
-|     |   ConsCdlAcc   |   ConsCdlPerfect    |   ImgCdlAcc   |   ImgCdlPerfect   |    BothPerfect   |
+## Performance of Diagram Formalizer on formalgeo7k test set
+| Model   |   ConsCdlAcc   |   ConsCdlPerfect    |   ImgCdlAcc   |   ImgCdlPerfect   |    BothPerfect   |
 |-----|----------------|---------------------|---------------|-------------------|------------------|
-|  siglip-0.4B-qwen2-0.5B  |    90.25         |      72.29           |     92.88        |  84.38 | 65.05  |
+|  Diagram Formalizer  |    90.25         |      72.29           |     92.88        |  84.38 | 65.05  |