%PDF-1.6
%ÐÔÅØ
15 0 obj
<<
/Length 5801
>>
stream
0 g 0 G
0 g 0 G
0 g 0 G
0 g 0 G
0 g 0 G
BT
/F47 11.9552 Tf 34.016 553.207 Td [(T)94(ext2LIVE:)-375(T)93(ext-Driv)32(en)-375(La)31(y)31(ered)-375(Image)-375(and)-375(Video)-375(Editing)]TJ/F8 9.9626 Tf 52.227 -29.888 Td [(Omer)-333(Bar-T)83(al)]TJ/F7 6.9738 Tf 61.076 3.615 Td [(1)]TJ/F13 6.9738 Tf 3.971 0 Td [(\003)]TJ/F8 9.9626 Tf 4.58 -3.615 Td [(,)-333(Dolev)-334(Ofri-Amar)]TJ/F7 6.9738 Tf 79.895 3.615 Td [(1)]TJ/F13 6.9738 Tf 3.972 0 Td [(\003)]TJ/F8 9.9626 Tf 4.58 -3.615 Td [(,)-333(Rafail)-334(F)84(ridman)]TJ/F7 6.9738 Tf 71.98 3.615 Td [(1)]TJ/F13 6.9738 Tf 3.971 0 Td [(\003)]TJ/F8 9.9626 Tf 4.58 -3.615 Td [(,)]TJ -184.198 -11.955 Td [(Y)83(oni)-333(Kasten)]TJ/F7 6.9738 Tf 53.743 3.615 Td [(2)]TJ/F8 9.9626 Tf 4.469 -3.615 Td [(,)-333(and)-334(T)84(ali)-334(Dek)28(el)]TJ/F7 6.9738 Tf 69.877 3.615 Td [(1)]TJ/F32 5.9776 Tf -186.851 -20.687 Td [(1)]TJ/F31 8.9664 Tf 7.223 -3.809 Td [(W)86(eizmann)-343(Institute)-342(of)-343(Science)]TJ/F32 5.9776 Tf 162.355 3.809 Td [(2)]TJ/F31 8.9664 Tf 7.223 -3.809 Td [(NVIDIA)-342(Researc)28(h)]TJ
0 g 0 G
ET
1 0 0 1 34.208 312.866 cm
q
.18446 0 0 .18446 0 0 cm
q
1 0 0 1 0 0 cm
/Im1 Do
Q
Q
1 0 0 1 -34.208 -312.866 cm
BT
/F53 8.9664 Tf 34.016 303.283 Td [(Fig.)-197(1.)]TJ/F31 8.9664 Tf 29.996 0 Td [(T)86(ext2LIVE)-235(p)-29(erforms)]TJ/F54 8.9664 Tf 84.947 0 Td [(semantic)]TJ/F31 8.9664 Tf 35.52 0 Td [(,)]TJ/F54 8.9664 Tf 4.668 0 Td [(lo)52(c)53(alize)52(d)]TJ/F31 8.9664 Tf 35.49 0 Td [(edits)-235(to)-235(real-w)28(orld)-235(images)-235(\050b\051,)-235(or)-235(videos)]TJ -190.621 -10.959 Td [(\050c\051.)-290(Our)-290(k)29(ey)-290(idea)-290(is)-290(to)-290(generate)-290(an)]TJ/F54 8.9664 Tf 138.407 0 Td [(e)52(dit)-318(layer)]TJ/F31 8.9664 Tf 37.816 0 Td [({R)29(GBA)-290(image)-290(represen)29(ting)-290(the)-290(target)-290(edit)]TJ -176.223 -10.959 Td [(when)-391(comp)-29(osited)-391(o)28(v)29(er)-392(the)-391(original)-392(input)-391(\050a\051.)-392(This)-391(allo)28(ws)-391(us)-392(to)-391(use)-392(text)-391(to)-392(guide)-391(not)]TJ 0 -10.959 Td [(only)-362(the)-362(\014nal)-362(comp)-28(osite,)-362(but)-362(also)-362(the)-362(edit)-362(la)28(y)29(er)-362(itself)-362(\050target)-362(text)-362(prompts)-362(are)-362(sho)29(wn)]TJ 0 -10.958 Td [(ab)-28(o)28(v)29(e)-338(eac)29(h)-338(image\051.)-337(Our)-338(edit)-337(la)28(y)29(ers)-338(are)-337(syn)28(thesized)-337(b)28(y)-337(training)-338(a)-337(generator)-338(on)-337(a)]TJ/F54 8.9664 Tf 323.244 0 Td [(single)]TJ/F31 8.9664 Tf -323.244 -10.959 Td [(input,)-342(without)-343(relying)-343(on)-342(user-pro)29(vided)-343(masks)-343(or)-342(a)-343(pre-trained)-342(generator.)]TJ
0 g 0 G
0 g 0 G
/F53 8.9664 Tf 28.346 -19.873 Td [(Abstract.)]TJ
0 g 0 G
/F31 8.9664 Tf 48.516 0 Td [(W)86(e)-384(presen)29(t)-384(a)-383(metho)-29(d)-383(f)-1(or)-383(zero-shot,)-384(text-driv)29(en)-384(app)-28(earance)]TJ -48.516 -10.959 Td [(manipulation)-478(in)-479(natural)-479(im)1(ages)-479(and)-479(videos.)-478(Giv)28(en)-478(an)-479(input)-478(image)-479(or)]TJ 0 -10.959 Td [(video)-437(and)-436(a)-437(target)-437(text)-437(prompt,)-436(our)-437(goal)-437(is)-437(to)-436(edit)-437(the)-437(app)-28(earance)-437(of)]TJ 0 -10.959 Td [(existing)-315(ob)-57(jects)-314(\050e.g.,)-315(ob)-57(ject's)-315(te)1(x)-1(t)1(ure\051)-315(or)-315(augmen)29(t)-315(the)-314(scene)-315(with)-314(visual)]TJ 0 -10.959 Td [(e\013ects)-251(\050e.)1(g.,)-251(smok)29(e,)-251(\014re\051)-251(in)-250(a)-251(seman)29(tically)-251(me)1(aningful)-251(manner.)-250(W)85(e)-250(train)-251(a)]TJ 0 -10.959 Td [(generator)-309(using)-309(an)]TJ/F54 8.9664 Tf 77.014 0 Td [(internal)-336(dataset)]TJ/F31 8.9664 Tf 66.265 0 Td [(of)-309(training)-309(examples,)-309(extracted)-309(from)]TJ -143.279 -10.959 Td [(a)-319(single)-319(input)-319(\050ima)1(ge)-319(or)-319(video)-319(and)-319(target)-319(text)-319(prompt\051,)-318(while)-319(lev)28(eraging)]TJ 0 -10.959 Td [(an)]TJ/F54 8.9664 Tf 12.555 0 Td [(external)]TJ/F31 8.9664 Tf 35.635 0 Td [(pre-trained)-315(CLIP)-315(mo)-29(del)-315(to)-315(establish)-316(our)-315(losses.)-315(Rather)-315(than)]TJ -48.19 -10.959 Td [(directly)-316(generating)-316(the)-315(edited)-316(output,)-316(our)-316(k)29(ey)-316(idea)-316(is)-316(to)-316(generate)-316(an)]TJ/F54 8.9664 Tf 274.785 0 Td [(e)52(dit)]TJ -274.785 -10.958 Td [(layer)]TJ/F31 8.9664 Tf 24.536 0 Td [(\050color+opacit)29(y\051)-438(that)-438(is)-439(comp)-28(osited)-438(o)28(v)29(er)-438(the)-438(original)-438(input.)-438(This)]TJ -24.536 -10.959 Td [(allo)29(ws)-352(us)-352(to)-351(constrain)-352(the)-352(generation)-351(pro)-29(cess)-351(and)-352(main)28(tai)1(n)-352(high)-352(\014delit)29(y)]TJ 0 -10.959 Td [(to)-287(the)-286(original)-287(input)-287(via)-286(no)28(v)29(el)-287(text-driv)29(en)-287(losses)-287(that)-286(are)-287(applied)-287(directly)]TJ 0 -10.959 Td [(to)-239(the)-239(edit)-239(la)29(y)29(er.)-239(Our)-239(metho)-29(d)-239(n)1(either)-239(relies)-239(on)-239(a)-239(pre-trained)-239(genera)1(tor)-239(nor)]TJ 0 -10.959 Td [(requires)-415(user-pro)29(vided)-415(edit)-415(masks.)-415(W)85(e)-415(demonstrate)-415(lo)-28(calized,)-415(seman)28(tic)]TJ 0 -10.959 Td [(edits)-454(on)-453(high-resolution)-454(natural)-453(images)-454(and)-454(videos)-453(across)-454(a)-454(v)57(a)1(riet)28(y)-454(of)]TJ 0 -10.959 Td [(ob)-57(jects)-343(and)-342(scenes.)-343(Pro)-57(ject)-342(page:)]TJ
0 1 0 0 k 0 1 0 0 K
/F54 8.9664 Tf 139.574 0 Td [(https://text2live.github.io/)]TJ
0 g 0 G
/F53 8.9664 Tf -139.574 -12.888 Td [(Keyw)33(ords:)]TJ/F31 8.9664 Tf 53.453 0 Td [(text-guided)-343(i)1(mage)-343(and)-343(video)-342(editing,)-343(app)-28(earance)-343(edi)1(ting,)-343(CLIP)]TJ
0 g 0 G
ET
q
1 0 0 1 34.016 44.477 cm
[]0 d 0 J 0.398 w 0 0 m 56.693 0 l S
Q
BT
/F31 8.9664 Tf 43.978 34.016 Td [(*)-343(Denotes)-342(equal)-343(con)29(tribution.)]TJ
0 g 0 G
0 g 0 G
0 g 0 G
ET
endstream
endobj
11 0 obj
<<
/Type /XObject
/Subtype /Form
/FormType 1
/PTEX.FileName (./figures/teaser_final_redo_small.pdf)
/PTEX.PageNumber 1
/PTEX.InfoDict 30 0 R
/BBox [0 0 1856.03 921.163]
/LastModified (D:20220405234626+03'00')/Group 25 0 R
/Resources <<
/ColorSpace <<
/CS0 31 0 R
>>/ExtGState <<
/GS0 32 0 R
>>/Font << /T1_0 33 0 R/TT0 34 0 R>>
/ProcSet [ /PDF /Text /ImageC ]
/XObject <<
/Im0 35 0 R
/Im1 36 0 R
/Im10 37 0 R
/Im11 38 0 R
/Im12 39 0 R
/Im13 38 0 R
/Im2 40 0 R
/Im3 41 0 R
/Im4 42 0 R
/Im5 43 0 R
/Im6 44 0 R
/Im7 45 0 R
/Im8 46 0 R
/Im9 47 0 R
>>>>
/Length 2922
/Filter /FlateDecode
>>
stream
HäW]o\7}¿¿BöÃÈ")RÔkbá}X È`óÐwb'nýØÎæïïѽw<x¼ã¢@`@+Q$ER'¯Þ¤ðêMHE·F
áÍ«
¾áäùp?|þ(gªpÌN5¬®¶1¦Äaåp5N]
o5&â0¿¤ð><Ñk
Rb®9o´-R,ÉçÊZ×=spåê7
?Цͺ-©n£kî,¾õç3+S$@®Cf=sw>¼
7ÃËá¤ÅzuøåÍ
á~u3,Íüòbhg?ü3-^ÙÌÃòz8zwôë»ããåoÃQhc(%,W2
aù~8:½>ûpÎß_>\Þ|÷_®îxjÓNªÖòïV»Î?¾B» Äô¿/ßßGu©#ÆìóúwvGÉ»Î.#LÅýû£/Á5Eüxzç P0v°!s40¦*²høç¡M¶]¬)Lò)¹Ì{T
Q3wrzÂë[hýé»=Y%V¶¢pO<b®S6`M{>¬¥