Update app.py

app.py

@@ -143,7 +143,8 @@ def analyze_and_explain(image, keywords, language, pdf_path):
     step2_prompt = (
         "You are helping to describe one bar chart in a scientific paper.\n\n"
         f"Rough visual caption from an image model: {raw_caption}\n"
-        f"Figure keywords from the user: {keywords}\n\n"
+        f"Figure keywords from the user: {keywords}\n"
+        f"Short paper context: {context_summary}\n\n"
         "Rewrite this into exactly ONE clear English sentence that starts with "
         "'A bar chart showing'. The sentence must:\n"
         "- say what is on the horizontal axis in generic terms "
@@ -151,7 +152,7 @@ def analyze_and_explain(image, keywords, language, pdf_path):
         "- say what is on the vertical axis in generic terms "
         "(e.g. relative bleaching, response intensity);\n"
         "- mention what main conditions or treatments are compared "
-        "(e.g. low vs high temperature);\n"
+        "(e.g. low vs high temperature or CO2);\n"
         "- briefly describe the overall trend "
         "(e.g. higher temperature causes stronger bleaching).\n\n"
         "Do NOT mention study duration, location, or sample size.\n"
@@ -164,59 +165,49 @@ def analyze_and_explain(image, keywords, language, pdf_path):
     if not step2_en.lower().startswith("a bar chart showing"):
         step2_en = "A bar chart showing " + step2_en.lstrip()
 
-    # ---- Step 3 + Step 5 合并：1 次 small T5 ----
-    context_part = (
-        f"Short paper context: {context_summary}\n\n"
-        if context_summary else ""
-    )
-
+    # ---- Step 3 + Step 5 合并：1 次 small T5，输出三行 ----
     multi_prompt = (
         "You are writing about one results figure in a scientific paper.\n\n"
-        f"{context_part}"
+        f"Short paper context: {context_summary}\n"
         f"Figure description: {step2_en}\n\n"
-        "Produce three parts:\n"
-        "1) A Results-style explanation (3-4 sentences) describing the trend.\n"
-        "2) A concise 1-2 sentence figure caption.\n"
-        "3) A short 3-4 sentence paragraph for the Results/Discussion section.\n\n"
-        "Write in formal academic English. Do NOT describe methods, duration, or location. "
-        "Do NOT invent exact numbers or statistics.\n\n"
-        "Separate the three parts with the markers:\n"
-        "<EXPL>\n"
-        "<CAP>\n"
-        "<SUM>\n\n"
-        "Output format:\n"
-        "<EXPL>\n"
-        "...explanation...\n"
-        "<CAP>\n"
-        "...caption...\n"
-        "<SUM>\n"
-        "...summary...\n"
+        "Write three parts, in formal academic English:\n"
+        "1) Explanation: 3-4 sentences describing what the bar chart shows.\n"
+        "2) Caption: 1-2 sentences suitable as a figure caption.\n"
+        "3) Summary: 3-4 sentences for the Results/Discussion section.\n\n"
+        "Rules:\n"
+        "- Focus on how the response changes across treatments or coral groups.\n"
+        "- Mention which treatment tends to have higher or lower values.\n"
+        "- Do NOT describe methods, duration, location, or sample size.\n"
+        "- Do NOT invent exact numbers or statistics.\n\n"
+        "Output EXACTLY in this format (English):\n"
+        "Explanation: <your 3-4 sentence explanation>\n"
+        "Caption: <your 1-2 sentence caption>\n"
+        "Summary: <your 3-4 sentence summary>\n"
+        "Do not add any extra labels or sections."
     )
 
     multi_out = text_pipe(multi_prompt)[0]["generated_text"]
 
-    expl_part = ""
-    cap_part = ""
-    sum_part = ""
-
-    if "<EXPL>" in multi_out:
-        _, rest = multi_out.split("<EXPL>", 1)
-    else:
-        rest = multi_out
-
-    if "<CAP>" in rest:
-        expl_part, rest2 = rest.split("<CAP>", 1)
-    else:
-        expl_part, rest2 = rest, ""
-
-    if "<SUM>" in rest2:
-        cap_part, sum_part = rest2.split("<SUM>", 1)
-    else:
-        cap_part, sum_part = rest2, ""
+    # 用正则把 Explanation / Caption / Summary 三段切出来
+    exp_match = re.search(
+        r"Explanation:\s*(.*?)(?:Caption:|Summary:|$)",
+        multi_out,
+        flags=re.S
+    )
+    cap_match = re.search(
+        r"Caption:\s*(.*?)(?:Summary:|$)",
+        multi_out,
+        flags=re.S
+    )
+    sum_match = re.search(
+        r"Summary:\s*(.*)$",
+        multi_out,
+        flags=re.S
+    )
 
-    explanation_en = dedup_sentences(expl_part)
-    caption_en = dedup_sentences(cap_part)
-    summary_en = dedup_sentences(sum_part)
+    explanation_en = dedup_sentences(exp_match.group(1).strip()) if exp_match else ""
+    caption_en = dedup_sentences(cap_match.group(1).strip()) if cap_match else ""
+    summary_en = dedup_sentences(sum_match.group(1).strip()) if sum_match else ""
 
     # ---- Step 4：图像增强 ----
     enhanced_img = enhance_image_simple(image)
@@ -273,7 +264,7 @@ with gr.Blocks() as demo:
             )
             pdf_in = gr.File(
                 label="上传论文 PDF（可选，用于结合上下文解释图表）",
-                type="filepath"   # ✅ 这里改成 filepath
+                type="filepath"   # 这里必须是 'filepath'
             )
             run_btn = gr.Button("分析并美化图表", variant="primary")