Update README.md

README.md

@@ -172,3 +172,5 @@ with open(file_path, 'r') as file:
         write_jsonl(raw, output_json)
 ```
 ## Examples
+Figure below shows the performances of our model on the lite version of the visual rating benchmark using different strategies for instruct prompts. As anticipated, the model not employing coarse-to-fine CoT yields lower performance, which indicates inherent challenges in directly predicting ratings. In contrast, our baseline with coarse-to-fine CoT performs better, especially on zero-shot datasets, illustrating the effectiveness of the coarse-to-fine CoT in enhancing robust and general thinking ability for visual rating by learning the ordinal regression nature.
+![Visualization results of coarse-to-fine CoT on different datasets.](./Visualization_results.png)