update instruction

eval_round_r1.db

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a1a533d4d5865529ddc5a733df7afb48b05c4704ead3df9f35cf1169e3807610
+size 65536

src/eval_server.py

@@ -151,20 +151,7 @@ def app_main(args) -> None:
 
             # Collapsible task guide
             with gr.Accordion('Task Instructions (Click to expand)', open=False):
-                gr.Markdown(
-                    '## Question 1: Relevance Score (1-5)\n'
-                    'Rate how relevant the icon is to the given label on a scale of 1 to 5:\n'
-                    '- **5**: Strongly related (highly consistent with the label, clear match)\n'
-                    '- **4**: Moderately related (clear partial connection, overall closer to related)\n'
-                    '- **3**: Neutral/uncertain (ambiguous, could be seen as either related or unrelated)\n'
-                    '- **2**: Weakly related (some minor or indirect connection, overall closer to unrelated)\n'
-                    '- **1**: Completely unrelated (no clear connection)\n'
-                    '## Question 2: Keyword Presence\n'
-                    'Check the 10 keywords and indicate which ones are **NOT** visible in the icon image:\n'
-                    '- Select **specific keywords** that are missing from the image, OR\n'
-                    '- Select **"ALL WORDS PRESENT"** if all 10 keywords are visible in the image\n'
-                    'You must select at least one option (either specific missing keywords or "ALL WORDS PRESENT")'
-                )
+                gr.Markdown(q1_instruction + '\n' + q2_instruction)
 
             # Relevance score
             label_md = gr.Markdown(visible=False)

src/instuction_md.py

@@ -7,7 +7,8 @@ Every time you start evaluating, input your user ID (e.g.: janeding) in the inpu
 
 q1_instruction = """
 ## Question 1: Relevance Score (1-5)
-Rate how relevant the icon is to the given label on a scale of 1 to 5:
+Rate how relevant the icon is to the given label on a scale of 1 to 5.
+### Your answer should be mainly based on your first intuition, but you may adjust it after viewing the label.
 - **5**: Strongly related (highly consistent with the label, clear match)
 - **4**: Moderately related (clear partial connection, overall closer to related)
 - **3**: Neutral/uncertain (ambiguous, could be seen as either related or unrelated)