pipeline.py

import cv2
import pdb
import json
import time
import shutil
import asyncio
import os, sys
import argparse
import subprocess
from os import path
from pdf2image import convert_from_path
from pathlib import Path

print("Initializing...")

from Paper2Video.src.slide_code_gen import latex_code_gen
from Paper2Video.src.wei_utils import get_agent_config
from posterbuilder import build_poster as build_poster
from posterbuilder.build_poster import IMAGES_DIR_NAME

ROOT_DIR = Path(__file__).resolve().parent
P2V_ASSETS = ROOT_DIR / "Paper2Video" / "assets" / "demo" / "latex_proj"
P2P_ROOT   = ROOT_DIR / "Paper2Poster"
PB_ROOT    = ROOT_DIR / "posterbuilder"
sys.path.append(str(P2P_ROOT))

def copy_folder(src_dir, dst_dir):
    src_dir = Path(src_dir)
    dst_dir = Path(dst_dir)
    if not src_dir.exists():
        raise FileNotFoundError(f"no such dir: {src_dir}")
    if dst_dir.exists():
        shutil.rmtree(dst_dir)
    shutil.copytree(src_dir, dst_dir)
    print(f"✅ Copied folder {src_dir} → {dst_dir}")

def copytree_overwrite(src: Path, dst: Path):
    if dst.exists():
        shutil.rmtree(dst)
    shutil.copytree(src, dst)

def safe_copy(src: Path, dst: Path):
    dst.parent.mkdir(parents=True, exist_ok=True)
    shutil.copy2(src, dst)

def str2list(s): 
    return [int(x) for x in s.split(',')]

def run_paper2poster_content_build():
    print("🧩 Step 1.5: Preparing Paper2Poster inputs & generating poster contents ...")

    src_pdf = ROOT_DIR / "input" / "paper.pdf"
    dst_pdf = P2P_ROOT / "input" / "paper" / "paper.pdf"
    dst_pdf.parent.mkdir(parents=True, exist_ok=True)
    safe_copy(src_pdf, dst_pdf)
    print(f"   📄 Copied paper: {src_pdf.relative_to(ROOT_DIR)} → {dst_pdf.relative_to(ROOT_DIR)}")

    cmd = [
        sys.executable, "-m", "PosterAgent.new_pipeline",
        f'--poster_path={dst_pdf.relative_to(P2P_ROOT)}',
        '--model_name_t=4o',
        '--model_name_v=4o',
        '--poster_width_inches=48',
        '--poster_height_inches=36'
    ]
    print("   ▶ Running: python -m PosterAgent.new_pipeline ...")
    subprocess.run(cmd, cwd=str(P2P_ROOT), check=True)
    print("   ✅ PosterAgent.new_pipeline finished.")

    tag_prefix = IMAGES_DIR_NAME.split("_images_and_tables")[0]
    src_raw_content = P2P_ROOT / "contents" / f"{tag_prefix}_paper_raw_content.json"
    src_tree_split  = P2P_ROOT / "tree_splits" / f"{tag_prefix}_paper_tree_split_0.json"
    src_images_json = P2P_ROOT / IMAGES_DIR_NAME / "paper_images.json"

    dst_contents_dir = PB_ROOT / "contents"
    dst_raw_content  = dst_contents_dir / "poster_content.json"
    dst_tree_split   = dst_contents_dir / "arrangement.json"
    dst_fig_caption  = dst_contents_dir / "figure_caption.json"

    dst_root_raw     = PB_ROOT / "poster_content.json"
    dst_root_tree    = PB_ROOT / "arrangement.json"
    dst_root_figcap  = PB_ROOT / "figure_caption.json"

    safe_copy(src_raw_content, dst_raw_content)
    safe_copy(src_tree_split,  dst_tree_split)
    safe_copy(src_images_json, dst_fig_caption)
    safe_copy(src_raw_content, dst_root_raw)
    safe_copy(src_tree_split,  dst_root_tree)
    safe_copy(src_images_json, dst_root_figcap)

    print("   📦 JSON copied & renamed.")
    print("   ✅ Step 1.5 done.\n")

if __name__ == '__main__':
    parser = argparse.ArgumentParser(description='Paper2Video Generation Pipeline')
    parser.add_argument('--result_dir', type=str, default='./result/zeyu')
    parser.add_argument('--model_name_t', type=str, default='gpt-4.1')
    parser.add_argument('--model_name_v', type=str, default='gpt-4.1')
    parser.add_argument('--paper_latex_root', type=str, default=str(P2V_ASSETS))
    parser.add_argument('--ref_text', type=str, default=None)
    parser.add_argument('--if_tree_search', type=bool, default=True)
    parser.add_argument('--beamer_templete_prompt', type=str, default=None)
    parser.add_argument('--stage', type=str, default='["0"]')
    parser.add_argument('--arxiv_url', type=str, default=None)
    parser.add_argument('--openai_key', type=str, required=True, help='Your OpenAI API key')
    parser.add_argument('--gemini_key', type=str, required=True, help='Your Gemini API key')

    args = parser.parse_args()
    print("start")

    # ✅ 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

    # 清空 output
    output_dir = ROOT_DIR / "output"
    if output_dir.exists():
        print(f"   🧹 Clearing old output directory: {output_dir.relative_to(ROOT_DIR)}")
        shutil.rmtree(output_dir)
    (output_dir / "latex_proj").mkdir(parents=True, exist_ok=True)
    (output_dir / "poster_latex_proj").mkdir(parents=True, exist_ok=True)
    (output_dir / "slide_imgs").mkdir(parents=True, exist_ok=True)
    print("   ✅ Created subfolders: latex_proj / poster_latex_proj / slide_imgs")

    # ================
    # Step 0: Download from arXiv
    # ================
    try:
        if args.arxiv_url:
            import requests, tarfile
            from io import BytesIO

            print(f"🧩 Step 0: Downloading from arXiv: {args.arxiv_url}")
            paper_id = args.arxiv_url.strip().split('/')[-1]
            input_dir = ROOT_DIR / "input"
            latex_proj_dir = input_dir / "latex_proj"

            if input_dir.exists():
                print(f"   🧹 Clearing old input directory: {input_dir.relative_to(ROOT_DIR)}")
                shutil.rmtree(input_dir)
            input_dir.mkdir(parents=True, exist_ok=True)
            latex_proj_dir.mkdir(parents=True, exist_ok=True)

            pdf_url = f"https://arxiv.org/pdf/{paper_id}.pdf"
            pdf_path = input_dir / "paper.pdf"
            print(f"   📄 Downloading PDF from {pdf_url} ...")
            r = requests.get(pdf_url)
            if r.status_code == 200:
                with open(pdf_path, 'wb') as f:
                    f.write(r.content)
                print(f"   ✅ Saved PDF → {pdf_path.relative_to(ROOT_DIR)}")
            else:
                raise RuntimeError(f"❌ Failed to download PDF (status {r.status_code})")

            src_url = f"https://arxiv.org/e-print/{paper_id}"
            print(f"   📦 Downloading LaTeX source from {src_url} ...")
            r = requests.get(src_url)
            if r.status_code == 200:
                try:
                    with tarfile.open(fileobj=BytesIO(r.content), mode="r:gz") as tar:
                        tar.extractall(path=latex_proj_dir)
                    print(f"   ✅ Extracted LaTeX source → {latex_proj_dir.relative_to(ROOT_DIR)}")
                except tarfile.ReadError:
                    print(f"   ⚠️  LaTeX source invalid, skipping.")
            else:
                print(f"   ⚠️  Failed to download LaTeX source.")
    except Exception as e:
        print(f"❌ Step 0 failed: {e}")

    # =========================
    # 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
    # =========================
    try:
        print("🧩 Step 2: Building poster ...")
        build_poster()
        print("✅ Step 2 done.")
    except Exception as e:
        print(f"❌ Step 2 failed: {e}")

    # =========================
    # Step 3: 导出 latex_proj
    # =========================
    try:
        src_lp = PB_ROOT / "latex_proj"
        dst_lp = ROOT_DIR / "output" / "poster_latex_proj"
        copytree_overwrite(src_lp, dst_lp)
        print(f"📦 Exported LaTeX project → {dst_lp.relative_to(ROOT_DIR)}")
    except Exception as e:
        print(f"❌ Step 3 failed: {e}")

    print("✅ Pipeline completed.")