Meta tags:
description=;

Headings (most frequently used words):

shape, aware, video, text, driven, layered, editing, interactive, demo, abstract, overview, comparison, with, baseline, methods, and, sota, application, interpolation, bibtex, acknowledgements, cvpr, 2023,

Text of the page (most frequently used words):
the (29), and (15), shape (11), video (10), editing (7), frame (7), aware (6), deformation (6), input (6), for (5), text (5), #driven (4), layered (4), swan (4), #method (4), with (4), atlas (4), mapping (4), nla (3), demo (3), jang (3), sports (3), black (3), object (3), can (3), methods (3), baseline (3), our (3), map (3), keyframe (3), per (3), field (3), edited (3), between (3), bar (2), text2live (2), lee (2), yao (2), chih (2), genevieve (2), chen (2), ting (2), qiu (2), elizabeth (2), huang (2), jia (2), bin (2), arxiv (2), 2023 (2), truck (2), jeep (2), duck (2), example (2), maps (2), interpolation (2), boat (2), edits (2), original (2), using (2), semantic (2), correspondence (2), vectors (2), into (2), each (2), first (2), pre (2), trained (2), output (2), videos (2), changes (2), compare (2), source (2), surf (2), thank, authors, yoni, kasten, dolev, ofri, oliver, wang, tali, dekel, omer, tal, rafail, fridman, providing, checkpoints, pretrained, models, acknowledgements, article, lee2023textvideoedit, title, author, journal, preprint, 2301, 13173, year, bibtex, train, white, linearly, interpolating, easily, interpolated, additional, without, application, single, multi, compared, dog, cat, yacht, ours, comparison, sota, module, back, via, pixel, are, transformed, space, same, consequently, samples, color, onto, deform, respecting, bases, decompose, unified, atlases, associated, dense, serves, through, bridge, overview, temporal, consistency, essential, applications, existing, work, representation, allows, propagating, consistently, these, however, only, edit, rather, than, due, limitation, fixed, texture, present, tackle, this, challenge, handle, propagate, all, frames, then, leverage, conditioned, diffusion, model, guidance, refining, distortion, completing, unseen, regions, experimental, results, demonstrate, that, achieve, consistent, favorably, state, art, appearance, abstract, rural, snow, grassland, background, suv, minivan, foreground, kid, kite, motorbike, libby, blackswan, car, turn, slide, left, right, interactive, code, paper, cvpr, university, maryland, college, park,

Text of the page (random words):
shape aware text driven layered video editing demo shape aware text driven layered video editing yao chih lee ji ze genevieve jang yi ting chen elizabeth qiu jia bin huang university of maryland college park cvpr 2023 paper code interactive video demo slide the bar to compare input left and output right input video car turn blackswan boat libby surf motorbike kite surf truck kid foreground source sports minivan suv background source grassland snow rural abstract temporal consistency is essential for video editing applications existing work on layered representation of videos allows propagating edits consistently to each frame these methods however can only edit object appearance rather than object shape changes due to the limitation of using a fixed uv mapping field for texture atlas we present a shape aware text driven video editing method to tackle this challenge to handle shape changes in video editing we first propagate the deformation field between the input and edited keyframe to all frames we then leverage a pre trained text conditioned diffusion model as guidance for refining shape distortion and completing unseen regions the experimental results demonstrate that our method can achieve shape aware consistent video editing and compare favorably with the state of the art overview our method first bases on a pre trained nla to decompose the video into unified atlases with the associated per frame uv mapping the dense semantic correspondence between the input and edited keyframe serves as the bridge between input and output videos for per frame deformation through the uv mapping and atlas our deformation module maps the edits back to the atlas via the original uv map by using the semantic correspondence of the keyframe the per pixel deformation vectors are transformed into the atlas space with the same uv mapping field consequently the uv map samples the color and the deformation vectors onto each frame to deform the original uv map respecting the edited shape comparison with baseline methods and sota ours example jeep sports black swan duck boat yacht dog cat compared method input multi frame baseline single frame baseline text2live application shape aware interpolation by linearly interpolating the deformation maps the object shape can be easily interpolated without additional frame interpolation methods example black swan duck black swan white swan jeep sports truck train bibtex article lee2023textvideoedit title shape aware text driven layered video editing demo author lee yao chih and jang ji ze genevieve jang and chen yi ting and qiu elizabeth and huang jia bin journal arxiv preprint arxiv 2301 13173 year 2023 acknowledgements we thank nla and text2live authors yoni kasten dolev ofri oliver wang tali dekel omer bar tal and rafail fridman for providing the checkpoints of nla pretrained models