update

app.py

@@ -139,22 +139,10 @@ def _prepare_workspace(logs):
 # ---------------------
 # Helpers for new features (post-processing)
 # ---------------------
-def _parse_rgb(s):
-    """Accepts '94,46,145' / '94 46 145' / '[94,46,145]' / '(94, 46, 145)' and returns (r,g,b) or None."""
-    if s is None:
-        return None
-    if isinstance(s, (tuple, list)) and len(s) == 3:
-        vals = s
-    else:
-        nums = re.findall(r"\d+", str(s))
-        if len(nums) < 3:
-            return None
-        vals = nums[:3]
+def _parse_rgb(hex):
     try:
-        r, g, b = (int(vals[0]), int(vals[1]), int(vals[2]))
-        if any(v < 0 or v > 255 for v in (r, g, b)):
-            return None
-        return (r, g, b)
+        hex = hex.lstrip('#')
+        return tuple(int(hex[i:i+2], 16) for i in (0, 2, 4))
     except Exception:
         return None
 
@@ -504,7 +492,10 @@ iface = gr.Interface(
             file_types=["image"],
         ),
         gr.File(label="🧩 Optional: Conference Logo (replaces right_logo.png)", file_count="single", file_types=["image"]),
-        gr.Textbox(label="🎨 Optional: Theme RGB (e.g., 94,46,145)", placeholder="94,46,145"),
+        gr.ColorPicker(
+            label="🎨 Theme Color (optional)",
+            value="#5E2E91"   # default purple (94,46,145)
+        ),
     ],
     outputs=[
         gr.Textbox(label="🧾 Logs (8~10 minutes)", lines=30, max_lines=50),
@@ -512,9 +503,7 @@ iface = gr.Interface(
     ],
     title="🎓 Paper2Poster",
     description="""
-    paper(https://arxiv.org/abs/2505.21497) | [GitHub](https://github.com/Paper2Poster/Paper2Poster) | project page (https://paper2poster.github.io/) 
-
-    # Paper2Poster
+    [paper](https://arxiv.org/abs/2505.21497) | [GitHub](https://github.com/Paper2Poster/Paper2Poster) | [project page] (https://paper2poster.github.io/) 
 
     Upload a paper, generate a poster for you.  
     Each paper takes approximately **8–10 minutes**.  

pipeline.py

@@ -21,9 +21,6 @@ P2P_ROOT   = ROOT_DIR / "Paper2Poster"
 PB_ROOT    = ROOT_DIR / "posterbuilder"
 sys.path.append(str(P2P_ROOT))
 
-print(f"🔒 Workspace ROOT_DIR = {ROOT_DIR}")
-print(f"🔒 This run is isolated under: {ROOT_DIR.resolve()}")
-
 def copy_folder(src_dir, dst_dir):
     src_dir = Path(src_dir)
     dst_dir = Path(dst_dir)
@@ -43,7 +40,7 @@ def safe_copy(src: Path, dst: Path):
     dst.parent.mkdir(parents=True, exist_ok=True)
     shutil.copy2(src, dst)
 
-def str2list(s):
+def str2list(s): 
     return [int(x) for x in s.split(',')]
 
 def run_paper2poster_content_build():
@@ -99,7 +96,13 @@ def _list_logo_files(logo_dir: Path):
     return files
 
 def _compose_logos_horizontally(logo_paths, out_path: Path, box_w=2000, box_h=476, gap=16):
-    # (same as your original; omitted comments for brevity)
+    """
+    宽度为硬约束：输出图像宽度必为 box_w（默认 2000px）。
+    多 logo 按比例统一缩放，拼接后刚好占满 box_w（包含间距）。
+    高度由比例自然决定，可能 < box_h，也可能 > box_h（甚至 > 2*box_h），不会再二次压缩。
+    透明背景，输出 PNG。
+    """
+    # 读取图片
     imgs = []
     for p in logo_paths:
         p = Path(p)
@@ -109,27 +112,35 @@ def _compose_logos_horizontally(logo_paths, out_path: Path, box_w=2000, box_h=47
     if n == 0:
         raise RuntimeError("No logo images found.")
 
+    # 原始总宽度（不含 gap）；拼接总宽 = sum(w_i) + gap*(n-1)
     widths  = [im.width for im in imgs]
     heights = [im.height for im in imgs]
     sum_w   = sum(widths)
     if sum_w <= 0:
         raise RuntimeError("All logo images have zero width.")
 
+    # 计算统一缩放比例，使：sum(w_i * s) + gap*(n-1) == box_w
+    # => s = (box_w - gap*(n-1)) / sum_w
     total_gap = max(0, gap * (n - 1))
     if box_w <= total_gap:
         raise ValueError(f"box_w({box_w}) too small vs total gaps({total_gap}). Increase box_w or reduce gap.")
     s = (box_w - total_gap) / float(sum_w)
 
+    # 按统一比例缩放（四舍五入到整数像素，避免累计误差）
     resized = []
     scaled_widths = []
+    scaled_heights = []
     for im, w, h in zip(imgs, widths, heights):
         nw = max(1, int(round(w * s)))
         nh = max(1, int(round(h * s)))
         resized.append(im.resize((nw, nh), Image.LANCZOS))
         scaled_widths.append(nw)
+        scaled_heights.append(nh)
 
+    # 由于整数取整，可能出现总宽 !=  box_w - total_gap；对若干图微调 1px 以精确对齐
     current_sum_w = sum(scaled_widths)
     diff = (box_w - total_gap) - current_sum_w
+    # 按从宽到窄/从大到小顺序均匀分配像素误差
     if diff != 0:
         order = sorted(range(n), key=lambda i: scaled_widths[i], reverse=(diff > 0))
         idx = 0
@@ -144,13 +155,14 @@ def _compose_logos_horizontally(logo_paths, out_path: Path, box_w=2000, box_h=47
                 remaining -= 1
             idx += 1
 
+    # 计算最终尺寸
     total_w = sum(scaled_widths) + total_gap
     assert total_w == box_w, f"width pack mismatch: got {total_w}, expect {box_w}"
     canvas_w = box_w
-    canvas_h = max(im.height for im in resized)
+    canvas_h = max(im.height for im in resized)  # 高度由比例自然决定（可能 > 2*box_h）
 
-    from PIL import Image as PILImage
-    canvas = PILImage.new("RGBA", (canvas_w, canvas_h), (0, 0, 0, 0))
+    # 画布 & 居中摆放（垂直方向居中）
+    canvas = Image.new("RGBA", (canvas_w, canvas_h), (0, 0, 0, 0))
     cur_x = 0
     for idx, im in enumerate(resized):
         y = (canvas_h - im.height) // 2
@@ -159,8 +171,13 @@ def _compose_logos_horizontally(logo_paths, out_path: Path, box_w=2000, box_h=47
         if idx != n - 1:
             cur_x += gap
 
+    # out_path.parent.mkdir(parents=True, exist_ok=True)
     canvas.save(out_path, format="PNG")
-    print(f"   🧩 Logos composed (width-locked) → {out_path.relative_to(ROOT_DIR)} (n={n}, final_size={canvas_w}x{canvas_h})")
+    print(f"   🧩 Logos composed (width-locked) → {out_path.relative_to(ROOT_DIR)} "
+          f"(n={n}, final_size={canvas_w}x{canvas_h})")
+
+
+
 
 if __name__ == '__main__':
     parser = argparse.ArgumentParser(description='Paper2Video Generation Pipeline')
@@ -179,11 +196,11 @@ if __name__ == '__main__':
     args = parser.parse_args()
     print("start")
 
-    # env
+    # ✅ 使用传入的 key 设置环境变量
     os.environ["OPENAI_API_KEY"] = args.openai_key
     os.environ["GEMINI_API_KEY"] = args.gemini_key
 
-    # clean and create run-local output
+    # 清空 output
     output_dir = ROOT_DIR / "output"
     if output_dir.exists():
         print(f"   🧹 Clearing old output directory: {output_dir.relative_to(ROOT_DIR)}")
@@ -193,7 +210,9 @@ if __name__ == '__main__':
     (output_dir / "slide_imgs").mkdir(parents=True, exist_ok=True)
     print("   ✅ Created subfolders: latex_proj / poster_latex_proj / slide_imgs")
 
-    # Step 0: arXiv download (same as your original)
+    # ================
+    # Step 0: Download from arXiv
+    # ================
     try:
         if args.arxiv_url:
             import requests, tarfile
@@ -236,13 +255,59 @@ if __name__ == '__main__':
     except Exception as e:
         print(f"❌ Step 0 failed: {e}")
 
-    # Step 1.5: content build
+    # =========================
+    # Step 1: Slide Generation
+    # =========================
+    # try:
+    #     print("🧩 Step 1: Generating Slides ...")
+    #     slide_latex_path = path.join(args.paper_latex_root, "slides.tex")
+    #     slide_image_dir = path.join(args.result_dir, 'slide_imgs')
+    #     os.makedirs(slide_image_dir, exist_ok=True)
+
+    #     start_time = time.time()
+    #     prompt_path = "./Paper2Video/src/prompts/slide_beamer_prompt.txt"
+
+    #     if args.if_tree_search:
+    #         usage_slide, beamer_path = latex_code_gen(
+    #             prompt_path=prompt_path,
+    #             tex_dir=args.paper_latex_root,
+    #             beamer_save_path=slide_latex_path,
+    #             model_config_ll=get_agent_config(args.model_name_t),
+    #             model_config_vl=get_agent_config(args.model_name_v),
+    #             beamer_temp_name=args.beamer_templete_prompt
+    #         )
+    #     else:
+    #         paper_latex_path = path.join(args.paper_latex_root, "main.tex")
+    #         usage_slide = latex_code_gen(
+    #             prompt_path=prompt_path,
+    #             tex_dir=args.paper_latex_root,
+    #             tex_path=paper_latex_path,
+    #             beamer_save_path=slide_latex_path,
+    #             model_config=get_agent_config(args.model_name_t)
+    #         )
+    #         beamer_path = slide_latex_path
+
+    #     if not os.path.exists(beamer_path):
+    #         raise FileNotFoundError(f"❌ Beamer PDF not found: {beamer_path}")
+
+    #     slide_imgs = convert_from_path(beamer_path, dpi=400)
+    #     for i, img in enumerate(slide_imgs):
+    #         img.save(path.join(slide_image_dir, f"{i+1}.png"))
+    #     print("✅ Step 1 done.")
+    # except Exception as e:
+    #     print(f"❌ Step 1 failed: {e}")
+
+    # =========================
+    # Step 1.5: Poster2Poster 内容生成
+    # =========================
     try:
         run_paper2poster_content_build()
     except Exception as e:
         print(f"❌ Step 1.5 failed: {e}")
 
-    # Step 2: build poster
+    # =========================
+    # Step 2: Build Poster
+    # =========================
     try:
         print("🧩 Step 2: Building poster ...")
         build_poster()
@@ -250,7 +315,9 @@ if __name__ == '__main__':
     except Exception as e:
         print(f"❌ Step 2 failed: {e}")
 
-    # Step 3: export latex & apply template & logos
+    # =========================
+    # Step 3: 导出 latex_proj & 处理 LOGO & 应用 template
+    # =========================
     try:
         src_lp = PB_ROOT / "latex_proj"
         dst_lp = ROOT_DIR / "output" / "poster_latex_proj"
@@ -277,13 +344,16 @@ if __name__ == '__main__':
             print("⚠️ template directory not found, skipping Step 3.5.")
 
         logos_out_dir = dst_lp / "logos"
+        # logos_out_dir.mkdir(parents=True, exist_ok=True)
         left_logo_path = logos_out_dir / "left_logo.png"
 
         if len(logo_files) == 1:
+            # 单图：拷贝并转成 PNG（以确保一致）
             im = Image.open(logo_files[0]).convert("RGBA")
             im.save(left_logo_path, format="PNG")
             print(f"🖼️  Single logo saved → {left_logo_path.relative_to(ROOT_DIR)}")
         else:
+            # 多图：拼接
             _compose_logos_horizontally(logo_files, left_logo_path, box_w=2000, box_h=476, gap=16)
 
         print("✅ Step 3 done.")

posterbuilder/convert.py

@@ -82,30 +82,39 @@ def fix_latex_escaped_commands(s: str) -> str:
     s = s.replace("\\}", "}")
     return s
 
+import re
+
 def escape_text(s: str) -> str:
     if not s:
         return ""
 
-    # ✅ 保护 $...$ 内的内容
-    math = []
-    def save_math(m):
-        math.append(m.group(0))
-        return f"__MATH{len(math)-1}__"
+    # ✅ 提取并暂存 math block（不产生 MATH 字符）
+    math_blocks = []
+
+    def store_math(m):
+        math_blocks.append(m.group(0))
+        # 用对象占位，避免出现标记文本
+        return f"\0{len(math_blocks)-1}\0"  # 不会出现在普通文本里
 
-    s = re.sub(r"\${1,2}.*?\${1,2}", save_math, s)
+    s = re.sub(r"\${1,2}.*?\${1,2}", store_math, s)
 
-    rep = {"&": r"\&", "%": r"\%", "$": r"\$", "#": r"\#", "_": r"\_",
-           "{": r"\{", "}": r"\}", "~": r"\textasciitilde{}", "^": r"\textasciicircum{}"}
+    # ✅ 执行普通字符转义
+    rep = {
+        "&": r"\&", "%": r"\%", "$": r"\$", "#": r"\#",
+        "_": r"\_", "{": r"\{", "}": r"\}",
+        "~": r"\textasciitilde{}", "^": r"\textasciicircum{}",
+    }
     for k, v in rep.items():
         s = s.replace(k, v)
 
-    # ✅ 恢复 math
-    for i, block in enumerate(math):
-        s = s.replace(f"__MATH{i}__", block)
+    # ✅ 恢复 math（无 MATH 字符产生）
+    for i, block in enumerate(math_blocks):
+        s = s.replace(f"\0{i}\0", block)
 
     return s
 
 
+
 def soft_wrap_title_for_logo(title: str, first_limit=68, next_limit=72) -> str:
     if not title or len(title) <= first_limit: return title
     def break_at(s: str, limit: int):