蔡易霖(Tsai, Yi-Lin)

AI專題,讓圖片動起來!AI topics, make pictures move!

By 蔡易霖(Tsai, Yi-Lin), August 01, 2020

31
0
2
AI專題,讓圖片動起來!
Translated by GoogleShow original (zh-TW)
本專案基於 Few-Shot Adversarial Learning of Realistic Neural Talking Head Models 進行實作。此模型分為兩個階段,第一個是meta-learning stage,另一個是fine-tuning stage。借助Few-shot Learning的概念,我們訓練一個Meta Learner,可以在相對極短的時間內針對少量樣本的臉部五官和圖片風格去做finetune,達成讓畫像動起來的驚人效果。通常在meta-learning stage訓練完後就可以生成圖,但會有不可識別身份的缺陷,再經過fine-tuning stage就可以改善表現不好的部分。 Demo: https://andy6804tw.github.io/talking-heads-web/material/svg-clip-path-hover-effect/dist/
This project is based on Few-Shot Adversarial Learning of Realistic Neural Talking Head Models. This model is divided into two stages, the first is the meta-learning stage, and the other is the fine-tuning stage. With the concept of Few-shot Learning, we train a Meta Learner that can finetune a small sample of facial features and picture styles in a relatively short period of time, and achieve amazing effects that make the portrait move. Usually the graph can be generated after the meta-learning stage is trained, but there will be the defect of unrecognizable identity, and then the fine-tuning stage can improve the poor performance. Demo: https://andy6804tw.github.io/talking-heads-web/material/svg-clip-path-hover-effect/dist/

Please login first.

Other works from 蔡易霖(Tsai,