"""Six-slide, evidence-backed status deck; reads frozen research outputs only."""
from pathlib import Path
import hashlib
import io
import json
import math
import shutil
import sys
import zipfile
import xml.etree.ElementTree as ET

PROJECT=Path(__file__).resolve().parents[1]
sys.path.insert(0,'/tmp/opencode/pallet_ppt_dependencies')
from PIL import Image,ImageDraw,ImageFont
from pptx import Presentation
from pptx.util import Inches,Pt
from pptx.dml.color import RGBColor
from pptx.oxml.xmlchemy import OxmlElement
from pptx.enum.text import MSO_ANCHOR
import cairosvg

OUT=PROJECT/'outputs/pallet_paper_status_ppt_20261115_v2'
DATA=PROJECT/'outputs/pallet_description_round8_training_v1'
FORMAL=PROJECT/'outputs/pallet_dataset_formal_v1/dataset'
NATIVE=PROJECT/'outputs/pallet_vlm_native_intensity_v1'
POSE=PROJECT/'outputs/pallet_pose_paper_benchmark_v5/report'
FONT=Path('/tmp/opencode/pallet_pose_design/NotoSansCJKsc-Regular.otf')
W,H=1600,900
INK='#182D43';MUTED='#5A6C7D';TEAL='#008A86';BLUE='#3478B7';RED='#C85B55';PALE='#F2F6F9';LINE='#DFE7ED'
provenance=[]


def read(path):return json.loads(path.read_text())
def rows(path):return [json.loads(s) for s in path.read_text().splitlines() if s]
def font(size):return ImageFont.truetype(str(FONT),size)
def sha(path):return hashlib.sha256(path.read_bytes()).hexdigest()
def wrap(text,size,width):
    result=[];line='';f=font(size)
    for c in text:
        if c=='\n':result.append(line);line='';continue
        if f.getlength(line+c)>width and line:result.append(line);line=c
        else:line+=c
    if line:result.append(line)
    return result
def text(im,xy,value,size=26,fill=INK,width=None,spacing=1.4):
    d=ImageDraw.Draw(im);lines=wrap(value,size,width) if width else value.split('\n')
    for i,line in enumerate(lines):d.text((xy[0],xy[1]+round(i*size*spacing)),line,font=font(size),fill=fill,anchor='lt')
    return len(lines)*size*spacing
def panel(title,w=480,h=510):
    im=Image.new('RGB',(w,h),PALE);d=ImageDraw.Draw(im);d.rounded_rectangle((0,0,w-1,h-1),radius=22,outline=LINE,width=2,fill=PALE)
    text(im,(25,24),title,28);d.line((25,70,w-25,70),fill=LINE,width=2);return im
def fit(im,src,box):
    if isinstance(src,(str,Path)):src=Image.open(src).convert('RGB')
    x,y,w,h=map(int,box);scale=min(w/src.width,h/src.height);res=src.resize((max(1,round(src.width*scale)),max(1,round(src.height*scale))),Image.Resampling.LANCZOS)
    im.paste(res,(x+(w-res.width)//2,y+(h-res.height)//2))
def source(path,label,key=None):
    provenance.append(dict(path=str(path.relative_to(PROJECT)) if path.is_relative_to(PROJECT) else str(path),sha256=sha(path),usage=label,target_key=key))
def save_asset(im,name):
    path=OUT/'assets'/f'{name}.png';im.save(path);return path
def crop_record(r,polygon=False,color=TEAL):
    path=FORMAL/r['source']['image'];source(path,'真实原生强度图/目标裁剪',r['target_key'])
    roi=r['roi'];im=Image.open(path).convert('RGB').crop(tuple(roi));scale=3
    im=im.resize((im.width*scale,im.height*scale),Image.Resampling.NEAREST)
    if polygon:
        d=ImageDraw.Draw(im);pts=[((x-roi[0])*scale,(y-roi[1])*scale) for x,y in r['polygon']]
        d.line(pts+[pts[0]],fill=color,width=4)
        for x,y in pts:d.ellipse((x-4,y-4,x+4,y+4),fill='#F8CC55')
    return im
def scene(sample,title,caption):
    im=panel(title);p=NATIVE/'sample/images'/f'{sample}.png';source(p,caption,sample);fit(im,p,(20,93,440,322));text(im,(25,441),caption,23,width=430);return im
def box(im,xywh,label,sub=None,color=TEAL):
    x,y,w,h=xywh;d=ImageDraw.Draw(im);d.rounded_rectangle((x,y,x+w,y+h),radius=14,fill='white',outline=color,width=2)
    f=font(27);d.text((x+w/2,y+19),label,font=f,fill=color,anchor='mt')
    if sub:text(im,(x+16,y+57),sub,21,MUTED,width=w-32)
def arrow(im,x,y,length=28,color=MUTED):
    d=ImageDraw.Draw(im);d.line((x,y,x,y+length),fill=color,width=3);d.polygon([(x,y+length+5),(x-7,y+length-4),(x+7,y+length-4)],fill=color)


def build_assets():
    training={r['target_key']:r for r in rows(DATA/'private/training.jsonl')}
    sem={r['target_key']:r for r in rows(DATA/'description_data/records.jsonl')}
    pred={r['target_key']:r for r in rows(DATA/'vlm_run_v1/component_branch/holdout_predictions.jsonl')}
    selected=rows(DATA/'private/release_selected.jsonl')
    assets={}
    assets['scene']=save_asset(scene('S003','典型托盘场景','原生强度图 · 真实采集'),'scene')
    assets['occlusion']=save_asset(scene('S083','叉齿遮挡场景','人工观察：插齿遮挡、右腿大部遮挡'),'occlusion')
    im=panel('论文主线：几何 × 语义')
    box(im,(42,100,396,92),'几何感知','类别、轮廓与点云位姿 · 已研究')
    arrow(im,240,201,24);box(im,(42,240,396,92),'轻量VLM','结构字段与场景描述 · 已训练')
    arrow(im,240,341,24);box(im,(42,380,396,94),'叉取可行性','几何约束融合与实机闭环 · 待验证',RED)
    assets['mainflow']=save_asset(im,'mainflow')
    im=panel('多形态数据覆盖')
    for i,cls in enumerate((1,2,6,12,13,41)):
        r=next(r for r in training.values() if r['class_id']==cls);x=20+(i%2)*235;y=90+(i//2)*133
        fit(im,crop_record(r),(x,y,220,92));text(im,(x,y+98),f'Class {cls}',20,MUTED)
    assets['diversity']=save_asset(im,'diversity')
    annotated=training['cb27610c0ab1aebabe0b48a6']
    im=panel('轮廓与顶点已人工复核');fit(im,crop_record(annotated),(20,100,440,140));fit(im,crop_record(annotated,True),(20,285,440,140))
    text(im,(25,248),'原图裁剪',21,MUTED);text(im,(25,445),'人工轮廓 / 角点 · '+str(annotated['K'])+'点',23,TEAL)
    assets['annotation']=save_asset(im,'annotation')
    r=next(r for r in training.values() if '遮挡' in r['description']['text'] and len(r['description']['text'])<45)
    im=panel('几何与语义联合记录');fit(im,crop_record(r,True),(20,100,440,180));text(im,(25,310),f"Class {r['class_id']}  ·  K={r['K']}",24,TEAL)
    text(im,(25,356),'人工描述：'+r['description']['text'],23,width=430)
    assets['semantic_label']=save_asset(im,'semantic_label')
    machine=next(r for r in selected if r['classified_class_id']==2)
    im=panel('分类／轮廓模型输出');fit(im,crop_record(machine),(20,95,440,150));fit(im,crop_record(machine,True,BLUE),(20,280,440,150))
    text(im,(25,249),'模型输入裁剪',21,MUTED);text(im,(25,447),f"预测Class {machine['classified_class_id']} · 严格K草稿",23,BLUE)
    assets['prediction']=save_asset(im,'prediction')
    key='0012bb42474000ca70805d48';p=POSE/'cases'/f'{key}.svg';source(p,'历史位姿研究：原始XYZ的XY投影',key)
    svg=ET.fromstring(p.read_bytes());labels=[]
    for elem in list(svg):
        if elem.tag.endswith('}text'):labels.append(dict(elem.attrib,value=elem.text or ''));svg.remove(elem)
        elif elem.tag.endswith('}circle') and elem.get('opacity'):elem.set('opacity','.75')
    cloud=Image.open(io.BytesIO(cairosvg.svg2png(bytestring=ET.tostring(svg),output_width=1200,output_height=740))).convert('RGB')
    cd=ImageDraw.Draw(cloud)
    for label in labels:
        if label['value'].startswith('equal scale'):continue
        x=float(label['x'])*2;y=float(label['y'])*2;anchor={'end':'rs','middle':'ms'}.get(label.get('text-anchor'),'ls')
        cd.text((x,y),label['value'],font=font(21),fill=MUTED,anchor=anchor)
    im=panel('点云位姿研究');fit(im,cloud,(12,104,456,310))
    text(im,(25,420),'参考 / 几何法 / 论文适配法',21,MUTED)
    text(im,(25,456),'实测XY投影 · 等比例 · 历史快照',21,MUTED)
    assets['pointcloud']=save_asset(im,'pointcloud')
    im=panel('同一托盘的双面候选')
    p=PROJECT/'outputs/pallet_pose_dimensions_v1/pair_contact_01.jpg';source(p,'历史双面几何配对P01，仅候选非尺寸真值')
    raw=Image.open(p).convert('RGB').crop((175,173,375,265));fit(im,raw,(20,105,440,280))
    text(im,(25,405),'双面结构与宽度关联已分析',24,TEAL);text(im,(25,450),'候选配对仍需逐例确认',22,MUTED)
    assets['faces']=save_asset(im,'faces')
    for label,key,title in [('vlm_good','cb27610c0ab1aebabe0b48a6','真实留出案例：字段一致'),('vlm_bad','2433a1eda849add0209a91ab','真实留出案例：材质错误')]:
        r=training[key];p=pred[key];s=sem[key];im=panel(title);fit(im,crop_record(r),(20,105,440,180))
        val=p['fields'];text(im,(25,315),'模型：'+f"{val['支腿']}支腿 / {val['横梁']}横梁 / {val['材质']}",23,TEAL if label=='vlm_good' else RED,width=430)
        val=s['target'];text(im,(25,362),'参考：'+f"{val['支腿']}支腿 / {val['横梁']}横梁 / {val['材质']}",23,INK,width=430)
        text(im,(25,434),'逐例展示，不代表总体性能',20,MUTED)
        assets[label]=save_asset(im,label)
    metrics=read(DATA/'vlm_run_v1/component_branch/holdout_metrics.json')
    im=panel('微调后结构字段一致率')
    for i,f in enumerate(('支腿','横梁','材质')):
        value=metrics['attribute_exact'][f]['accuracy']*100;y=115+i*102;text(im,(25,y),f,26);text(im,(326,y),f'{value:.2f}%',26,TEAL)
        d=ImageDraw.Draw(im);d.rounded_rectangle((25,y+44,455,y+64),radius=10,fill=LINE);d.rounded_rectangle((25,y+44,25+4.3*value,y+64),radius=10,fill=TEAL)
    text(im,(25,447),'组件留出 n=1,951 · 非新盲测',21,MUTED)
    assets['vlm_metrics']=save_asset(im,'vlm_metrics')
    def paired_chart(title,labels,old,new,name):
        im=panel(title,740,510);text(im,(28,91),'第七轮',22,BLUE);text(im,(150,91),'第八轮',22,TEAL)
        d=ImageDraw.Draw(im);left=190;right=650;top=166;bottom=402
        for tick in (85,90,95,100):
            x=left+(tick-85)/15*(right-left);d.line((x,top-10,x,bottom),fill=LINE,width=2);text(im,(x-17,415),str(tick),21,MUTED)
        for i,(label,a,b) in enumerate(zip(labels,old,new)):
            y=top+i*92;text(im,(28,y+5),label,24)
            xa=left+(a-85)/15*(right-left);xb=left+(b-85)/15*(right-left)
            d.line((min(xa,xb),y+20,max(xa,xb),y+20),fill='#A9B9C8',width=4)
            d.ellipse((xa-7,y+13,xa+7,y+27),fill=BLUE);d.ellipse((xb-7,y+13,xb+7,y+27),fill=TEAL)
            text(im,(min(min(xa,xb)-22,495),y-22),f'{a:.2f} → {b:.2f}',23,RED if b<a else TEAL)
        text(im,(28,470),'同一份最新标签对比；横轴85%–100%',21,MUTED)
        return save_asset(im,name)
    oldm=read(DATA/'mask_run_v1/old_component_branch_comparison.json');newm=read(DATA/'mask_run_v1/independent_branch/holdout_metrics.json')
    assets['mask_compare']=paired_chart('分类回退，轮廓略升',['细类准确率','大类准确率','IoU × 100'],[oldm['fine_top1']*100,oldm['parent_top1']*100,oldm['segmentation']['iou_median']*100],[newm['fine_top1']*100,newm['parent_top1']*100,newm['segmentation']['iou_median']*100],'mask_compare')
    oldv=read(DATA/'vlm_run_v1/old_component_comparison/holdout_metrics.json')
    assets['vlm_compare']=paired_chart('材质改善，支腿／横梁回退',['支腿一致率','横梁一致率','材质一致率'],[oldv['attribute_exact'][f]['accuracy']*100 for f in ('支腿','横梁','材质')],[metrics['attribute_exact'][f]['accuracy']*100 for f in ('支腿','横梁','材质')],'vlm_compare')
    im=panel('① 补齐决策标注');p=NATIVE/'sample/images/S090.png';source(p,'下一步评测案例：已标注左腿出视野','S090');fit(im,p,(25,95,430,255))
    text(im,(25,375),'可叉取 / 不可叉取 / 无法判断',24,TEAL,width=430);text(im,(25,427),'拟标注：证据区域与判断依据',22,MUTED,width=430)
    assets['next_labels']=save_asset(im,'next_labels')
    im=panel('② 融合几何与语义')
    box(im,(35,105,410,95),'强度图＋深度／点云','可见结构、距离与孔洞几何')
    arrow(im,240,211,22);box(im,(35,250,410,95),'几何约束＋VLM描述','区分视觉证据与推断')
    arrow(im,240,356,22);box(im,(35,395,410,82),'可解释的叉取可行性',None,RED)
    assets['next_fusion']=save_asset(im,'next_fusion')
    im=panel('③ 完成关键实验')
    box(im,(35,105,410,95),'融合消融','几何 / VLM / 融合三组对照')
    arrow(im,240,211,22);box(im,(35,250,410,95),'未见类型与复杂场景','零样本／少样本仍待验证')
    arrow(im,240,356,22);box(im,(35,395,410,82),'实机叉取与失败分析',None,RED)
    assets['next_tests']=save_asset(im,'next_tests')
    return assets


def build():
    if OUT.exists():raise RuntimeError('Use a fresh output version; preserve existing deck')
    (OUT/'assets').mkdir(parents=True);(OUT/'previews').mkdir()
    assets=build_assets()
    paragraphs=[
        '面向遮挡、出视野与复杂结构，探索“几何感知＋轻量VLM”的托盘叉取可行性判断。当前已完成数据建设、结构描述与几何模型研究；几何—语义融合和实机决策闭环仍待验证。',
        '已冻结8,698个目标、8,659张原图，覆盖40细类与9大类，形成类别、轮廓顶点和语义描述的联合复核流程。训练／组件留出为6,679／2,019项，按关联组件划分，避免同图跨集合。',
        '已完成分类／轮廓模型迭代，以及点云几何、论文方法适配和PointNeXt位姿对照。最新细类准确率89.80%，轮廓IoU中位数0.8911；历史位姿研究可输出中心与方向，但尚非实机叉取验证。',
        'Qwen3.5-0.8B经LoRA微调后，支腿、横梁、材质一致率分别为95.64%、97.74%、95.23%，JSON有效率100%（组件留出1,951项）。结构描述已有基础，场景风险和叉取可行性仍待可靠评测。',
        '第八轮较第七轮，轮廓与材质改善，分类、支腿和横梁指标回退，尚无全面提升。长尾类别、未见形态与强度图缺少深度仍是瓶颈；现有反复迭代留出结果不等于新场景或零样本泛化能力。',
        '下一步围绕“可叉取／不可叉取／无法判断”构建带证据的评测集，补充遮挡、出视野、多面选择和孔洞可见性标注；引入深度与几何约束，通过融合消融、未见类别及实机实验验证实际收益。'
    ]
    titles=['研究目标：从托盘识别到叉取可行性','数据基础：几何与语义联合标注','几何感知：轮廓学习与位姿研究','轻量VLM：结构描述已有基础','实验结论：局部改善，仍有瓶颈','下一步：验证几何与语义融合价值']
    names=[['scene','occlusion','mainflow'],['diversity','annotation','semantic_label'],['prediction','pointcloud','faces'],['vlm_good','vlm_bad','vlm_metrics'],['mask_compare','vlm_compare'],['next_labels','next_fusion','next_tests']]
    note_sources=[['用户附件：托盘识别+VLM论文.txt','docs/pallet_vlm_native_intensity_v1.md'],['outputs/pallet_description_round8_training_v1/preparation_report.json','outputs/pallet_description_round8_training_v1/source_replay.json'],['outputs/pallet_description_round8_training_v1/mask_run_v1/independent_branch/holdout_metrics.json','outputs/pallet_pose_paper_benchmark_v5/REPORT_zh.md','docs/pallet_pose_dimensions_v1.md'],['outputs/pallet_description_round8_training_v1/vlm_run_v1/component_branch/holdout_metrics.json','outputs/pallet_description_round8_training_v1/description_data/records.jsonl'],['outputs/pallet_description_round8_training_v1/TRAINING_REPORT.md'],['用户附件：托盘识别+VLM论文.txt']]
    for path in [DATA/'preparation_report.json',DATA/'TRAINING_REPORT.md',DATA/'mask_run_v1/independent_branch/holdout_metrics.json',DATA/'vlm_run_v1/component_branch/holdout_metrics.json',DATA/'description_data/records.jsonl',DATA/'vlm_run_v1/component_branch/holdout_predictions.jsonl']:
        source(path,'数值或案例引用')
    prs=Presentation();prs.slide_width=Inches(W/120);prs.slide_height=Inches(H/120)
    prs.core_properties.title='托盘识别与轻量VLM｜论文工作现状';prs.core_properties.subject='第八轮冻结数据与已有几何研究；已完成与待验证分开';prs.core_properties.author='Research status';prs.core_properties.keywords='托盘, VLM, 几何感知, 工作现状'
    previews=[];slide_specs=[]
    def add_text(slide,x,y,w,h,lines,size,color):
        shape=slide.shapes.add_textbox(Inches(x/120), Inches(y/120), Inches(w/120), Inches(h/120));tf=shape.text_frame
        tf.margin_left=tf.margin_right=tf.margin_top=tf.margin_bottom=0;tf.word_wrap=False;tf.vertical_anchor=MSO_ANCHOR.TOP
        p=tf.paragraphs[0];p.line_spacing=Pt(size*1.42*72/120);p.space_before=Pt(0);p.space_after=Pt(0)
        for i,line in enumerate(lines):
            if i:p.add_line_break()
            r=p.add_run();r.text=line;r.font.name='Noto Sans CJK SC';r.font.size=Pt(size*72/120);r.font.color.rgb=RGBColor.from_string(color.lstrip('#'))
            props=r._r.get_or_add_rPr();ea=OxmlElement('a:ea');ea.set('typeface','Noto Sans CJK SC');props.append(ea)
        return shape
    for i,(title,para,figs) in enumerate(zip(titles,paragraphs,names)):
        slide=prs.slides.add_slide(prs.slide_layouts[6]);im=Image.new('RGB',(W,H),'white');d=ImageDraw.Draw(im)
        # Only title, a single short paragraph, and 2–3 figure images on each slide.
        titlelines=[title];bodylines=wrap(para,30,1480)
        assert len(bodylines)<=3 and font(49).getlength(title)<1480
        text(im,(60,57),title,49);text(im,(60,151),'\n'.join(bodylines),30,MUTED,spacing=1.42)
        add_text(slide,60,52,1480,78,titlelines,49,INK);add_text(slide,60,144,1480,145,bodylines,30,MUTED)
        positions=[(60+500*j,320,480,510) for j in range(3)] if len(figs)==3 else [(60,320,740,510),(820,320,740,510)]
        for name,(x,y,w,h) in zip(figs,positions):
            p=assets[name];im.paste(Image.open(p),(x,y));slide.shapes.add_picture(str(p),Inches(x/120), Inches(y/120),width=Inches(w/120),height=Inches(h/120))
        notes='第'+str(i+1)+'页｜资料截至2026-11-15；最新模型数值来自第八轮冻结实验。\n来源：\n'+'\n'.join(note_sources[i])+'\n图示使用真实既有原生强度图、轮廓或点云；流程图表示研究思路。选例为定性展示，非随机代表。'
        if i==2:notes+='\n轮廓模型草稿与历史位姿、双面配对来自不同快照，不能合并为已打通的端到端系统。位姿结果不是实机成功率或独立物理精度。'
        if i==3:notes+='\n案例一致/错误仅就支腿、横梁、材质三个监督字段判定；不代表完整观察描述正确。原始强度图并非自然RGB。'
        if i==4:notes+='\n所有数值为旧/新模型在同一份最新标签的配对对比。图中IoU乘100作百分比标度。训练侧缺Class56、128。'
        slide.notes_slide.notes_text_frame.text=notes
        preview=OUT/'previews'/f'slide_{i+1:02d}.png';im.save(preview);previews.append(im)
        slide_specs.append(dict(page=i+1,title=title,paragraph=para,figures=figs,sources=note_sources[i],layout='title + one paragraph + images',preview_kind='same-layout Pillow rendering, not Office screenshot'))
    ppt=OUT/'托盘识别与VLM_论文工作现状_6页.pptx';prs.save(ppt)
    # A font-independent same-layout PDF and PNG handout accompanies editable text.
    previews[0].save(OUT/'托盘识别与VLM_论文工作现状_预览.pdf',save_all=True,append_images=previews[1:],resolution=120)
    contact=Image.new('RGB',(1640,1450),'#E6ECF0')
    for i,im in enumerate(previews):contact.paste(im.resize((800,450),Image.Resampling.LANCZOS),(10+(i%2)*820,10+(i//2)*480))
    contact.save(OUT/'六页总览.png')
    (OUT/'sources.json').write_text(json.dumps(dict(slides=slide_specs,assets=provenance),ensure_ascii=False,indent=2))
    (OUT/'字体').mkdir();shutil.copyfile(FONT,OUT/'字体/NotoSansCJKsc-Regular.otf')
    license_path=PROJECT/'outputs/pallet_pose_dimensions_v1/FONT_LICENSE.txt'
    if license_path.exists():shutil.copyfile(license_path,OUT/'字体/FONT_LICENSE.txt')
    (OUT/'说明.txt').write_text('六页极简工作现状：每页仅标题、一段短文、2–3张图。\nPPT标题和正文可编辑，配图为独立高清PNG。字体：Noto Sans CJK SC，配套字体与许可证在字体文件夹。\nPDF/PNG为同一版式定义直接生成的预览，不是Office软件截图；无需安装字体即可阅读。\n科研来源放在每页备注与sources.json中。当前输入为原生强度图，不冒充RGB；叉取可行性、zero/few-shot与实机闭环均标为待验证。\n')
    # Structural acceptance: exact number of slides, native text shapes and pictures.
    loaded=Presentation(ppt);assert len(loaded.slides)==6
    from pptx.enum.shapes import MSO_SHAPE_TYPE
    checks=[]
    for i,s in enumerate(loaded.slides):
        texts=[sh.text for sh in s.shapes if sh.has_text_frame];pics=[sh for sh in s.shapes if sh.shape_type==MSO_SHAPE_TYPE.PICTURE]
        assert len(texts)==2 and len(pics)==len(names[i]) and texts[0]==titles[i]
        assert texts[1].replace('\v','').replace('\n','')==paragraphs[i]
        assert all(sh.left>=0 and sh.top>=0 and sh.left+sh.width<=loaded.slide_width and sh.top+sh.height<=loaded.slide_height for sh in s.shapes)
        checks.append(dict(page=i+1,native_text_shapes=2,pictures=len(pics),within_canvas=True,notes_sources_present=True))
    from fontTools.ttLib import TTFont
    with TTFont(FONT) as f:cmap=f.getBestCmap()
    assert not {ord(c) for c in ''.join(titles+paragraphs) if '\u3400'<=c<='\u9fff'}-set(cmap)
    with zipfile.ZipFile(ppt) as z:assert z.testzip() is None
    (OUT/'verification.json').write_text(json.dumps(dict(state='six_slide_structure_and_sources_verified',slides=checks,pptx_sha256=sha(ppt),native_office_render_performed=False,preview_source='same layout renderer',production_writes=0),ensure_ascii=False,indent=2))
    with zipfile.ZipFile(OUT/'论文工作现状_完整包.zip','w',zipfile.ZIP_DEFLATED) as archive:
        for p in OUT.rglob('*'):
            if p.is_file() and p.suffix!='.zip':archive.write(p,p.relative_to(OUT))
    print(ppt);print(OUT/'六页总览.png')


if __name__=='__main__':build()
