Post

๐ŸŽญ๐ŸŽญ FaceLift new SOTA in 2D Landmarks ๐ŸŽญ๐ŸŽญ

๐Ÿ‘‰Flawless AI unveils a novel semi-supervised approach that learns 3D landmarks by directly lifting (visible) hand-labeled 2D landmarks and ensures better definition alignment, with no need for 3D landmark datasets.

๐‡๐ข๐ ๐ก๐ฅ๐ข๐ ๐ก๐ญ๐ฌ:

  • โœ…Novel semi-supervised HQ 2D landmarks
  • โœ…3D-aware GAN prior to tackle 2D-3D lifting
  • โœ…Accurate 3D landmarks aligned with 2D GT
  • โœ…3D ViT leveraging volumetric consistency
  • โœ…Novel SOTA on 2D-3D faces datasets

๐Ÿ‘‰Discussion https://t.me/AI_DeepLearning

๐Ÿง™Paper Authors: David Ferman Pablo Garrido Gaurav Bharaj Flawless AI

Translate to Korean

๐Ÿ‘‰Flawless AI๋Š” ์†์œผ๋กœ ๋ผ๋ฒจ๋งํ•œ 2D ๋žœ๋“œ๋งˆํฌ๋ฅผ ์ง์ ‘ ๋“ค์–ด ์˜ฌ๋ ค 3D ๋žœ๋“œ๋งˆํฌ๋ฅผ ํ•™์Šตํ•˜๊ณ  3D ๋žœ๋“œ๋งˆํฌ ๋ฐ์ดํ„ฐ ์„ธํŠธ ์—†์ด ๋” ๋‚˜์€ ์„ ๋ช…๋„ ์ •๋ ฌ์„ ๋ณด์žฅํ•˜๋Š” ์ƒˆ๋กœ์šด ๋ฐ˜์ง€๋„ ์ ‘๊ทผ ๋ฐฉ์‹์„ ๊ณต๊ฐœํ•ฉ๋‹ˆ๋‹ค.

๋ฐœํ‘œ๋œ๐Ÿฅน ์ฝ”๋“œ ์—†์Œ

ํ•˜์ด๋ผ์ดํŠธ:

  • โœ…์ƒˆ๋กœ์šด ์ค€๊ฐ๋… HQ 2D ๋žœ๋“œ๋งˆํฌ
  • โœ…2D-3D ๋ฆฌํ”„ํŒ…์„ ๋‹ค๋ฃจ๊ธฐ ์ „ 3D ์ธ์‹ GAN
  • โœ…2D GT์— ๋งž์ถฐ ์ •๋ ฌ๋œ ์ •ํ™•ํ•œ 3D ๋žœ๋“œ๋งˆํฌ
  • โœ…์ฒด์  ์ผ๊ด€์„ฑ์„ ํ™œ์šฉํ•˜๋Š” 3D ViT
  • โœ…2D-3D ์–ผ๊ตด ๋ฐ์ดํ„ฐ ์„ธํŠธ์— ๋Œ€ํ•œ ์ƒˆ๋กœ์šด SOTA
This post is licensed under CC BY 4.0 by the author.