update explanations on MM-UPD Bench

constants.py

@@ -35,40 +35,13 @@ LEADERBORAD_INTRODUCTION = """
 <a href='https://arxiv.org/abs/2403.20331'><img src='https://img.shields.io/badge/cs.CV-Paper-b31b1b?logo=arxiv&logoColor=red'></a>
 </div>
 
-## What is MM-UPD Bench?
 MM-UPD Bench: A Comprehensive Benchmark for Evaluating the Trustworthiness of Vision Language Models (VLMs) in the Context of Unsolvable Problem Detection (UPD)
 
 Our MM-UPD Bench encompasses three benchmarks: MM-AAD, MM-IASD, and MM-IVQD.
 
-1\. **MM-AAD:** Benchmark for Absent Answer Detection (AAD).  
-MM-AAD Bench is a dataset where the correct answer option for each question is removed.  
-MM-AAD tests the model's capability to recognize when the correct answer is absent from the provided choices.
+Through these benchmarks, we aim to provide a comprehensive evaluation of VLMs across multiple senarios.
 
-2\. **MM-IASD:** Benchmark for Incompatible Answer Set Detection (IASD).  
-MM-IASD Bench is a dataset where the answer set is completely incompatible with the context specified by the question and the image.  
-MM-IASD tests the model's capability to recognize when the answer set is incompatible with the context.
-
-3\. **MM-IVQD:** Benchmark for Incompatible Visual Question Detection (IVQD).  
-MM-IVQD Bench is a dataset where the question is incompatible with the image.  
-MM-IVQD evaluates the VLMs' capability to discern when a question and image are irrelevant or inappropriate.
-
-
-## Characteristics of MM-UPD Bench
-We design MM-UPD Bench to provide a comprehensive evaluation of VLMs across multiple senarios.
-
-1\. **Multiple Senario Evaluation:** We carefully design prompts choices and examine the three senario: (i) Base (w/o instruction), (ii) Option (w/ additional option), (iii) Instruction (w/ additional instruction).
-
-2\. **Ability-Wise Evaluation:** We carefully decompose each benchmark into various abilities to reveal individual model's strengths and weaknesses.
-
-
-## About Evaluation Metrics
-We evaluate the performance of VLMs on MM-UPD Bench using the following metrics:
-1. **Dual accuracy:** The accuracy on standard-UPD pairs, where we count
-success only if the model is correct on both the standard and UPD questions.
-
-2. **Standard accuracy:** The accuracy on standard questions.
-
-3. **UPD (AAD/IASD/IVQD) accuracy:** The accuracy of AAD/IASD/IVQD questions. 
+As for more detail information, we can refer to 'About' section.
 
 
 Please follow the instructions in [UPD](https://github.com/AtsuMiyai/UPD) to upload the generated JSON file here. After clicking the `Submit Eval` button, click the `Refresh` button.
@@ -106,10 +79,42 @@ SUBMIT_INTRODUCTION = """# Submit on MM-UPD Benchmark Introduction
 
 
 LEADERBORAD_INFO = """
-      MM-UPD Bench is a comprehensive benchmark for evaluating the trustworthiness of Vision Language Models 
-      (VLMs) in the context of Unsolvable Problem Detection (UPD). MM-UPD encompasses three benchmarks: 
-      MM-AAD, MM-IASD, and MM-IVQD. Each benchmark cover a wide range of abilities. Through these benchmarks, 
-      we aim to provide a comprehensive evaluation of VLMs across multiple senarios.
+## What is MM-UPD Bench?
+MM-UPD Bench: A Comprehensive Benchmark for Evaluating the Trustworthiness of Vision Language Models (VLMs) in the Context of Unsolvable Problem Detection (UPD)
+
+Our MM-UPD Bench encompasses three benchmarks: MM-AAD, MM-IASD, and MM-IVQD.
+
+1\. **MM-AAD:** Benchmark for Absent Answer Detection (AAD).  
+MM-AAD Bench is a dataset where the correct answer option for each question is removed.  
+MM-AAD tests the model's capability to recognize when the correct answer is absent from the provided choices.
+
+2\. **MM-IASD:** Benchmark for Incompatible Answer Set Detection (IASD).  
+MM-IASD Bench is a dataset where the answer set is completely incompatible with the context specified by the question and the image.  
+MM-IASD tests the model's capability to recognize when the answer set is incompatible with the context.
+
+3\. **MM-IVQD:** Benchmark for Incompatible Visual Question Detection (IVQD).  
+MM-IVQD Bench is a dataset where the question is incompatible with the image.  
+MM-IVQD evaluates the VLMs' capability to discern when a question and image are irrelevant or inappropriate.
+
+
+## Characteristics of MM-UPD Bench
+We design MM-UPD Bench to provide a comprehensive evaluation of VLMs across multiple senarios.
+
+1\. **Multiple Senario Evaluation:** We carefully design prompts choices and examine the three senario: (i) Base (w/o instruction), (ii) Option (w/ additional option), (iii) Instruction (w/ additional instruction).
+
+2\. **Ability-Wise Evaluation:** We carefully decompose each benchmark into various abilities to reveal individual model's strengths and weaknesses.
+
+
+## About Evaluation Metrics
+We evaluate the performance of VLMs on MM-UPD Bench using the following metrics:
+1. **Dual accuracy:** The accuracy on standard-UPD pairs, where we count
+success only if the model is correct on both the standard and UPD questions.
+
+2. **Standard accuracy:** The accuracy on standard questions.
+
+3. **UPD (AAD/IASD/IVQD) accuracy:** The accuracy of AAD/IASD/IVQD questions. 
+
+
 """