AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Understanding Pictograph with Facial Features: End-to-End Sentence-Level Lip Reading of Chinese

Xiaobing Zhang, Haigang Gong, Xili Dai, Fan Yang, Nianbo Liu, Ming Liu

AAAI 2019 · Architectures and efficiency › RNNs and normalization · approach: Non-autoregressive translation · 39 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
replace WLAS with LipCH-Netlip reading · CCTVsentence-level accuracy: 36.7% -> 50.2%Improved
end-to-end training of Picture-to-Pinyin (ablation)VGG-M+ResNet+2-LSTM+CTC · CCTVPAR: 44.2% -> 46.3%Improved
use CTC instead of softmax loss (ablation)Picture-to-Pinyin · CCTVPAR: 29.7% -> 44.2%Improved
use ResNet instead of 6-layer DNN (ablation)Picture-to-Pinyin · CCTVPAR: 36.7% -> 41.2%Improved
use 3D conv instead of 2D (VGG-M) (ablation)Picture-to-Pinyin · CCTVPAR: 41.2% -> 35.1%Hurt
use bidirectional LSTM instead of unidirectional (ablation)Picture-to-Pinyin · CCTVPAR: 44.2% -> 41.4%Hurt
use GRU instead of LSTM in Picture-to-Pinyin (ablation)Picture-to-Pinyin · CCTVPAR: 44.2% -> 34.8%Hurt
add auxiliary data pretraining (ablation)Pinyin-to-Hanzi · CCTVHAR: 35.4% -> 39.8%Improved
add scheduled sampling (ablation)Pinyin-to-Hanzi · CCTVHAR: 39.8% -> 40.2%Improved
add decoder attention (ablation)Pinyin-to-Hanzi · CCTVHAR: 40.2% -> 45.7%Improved
add curriculum learning (ablation)LipCH-Net · CCTVHAR: 42.7% -> 43.1%Improved
add encoder attention for joint training (ablation)LipCH-Net · CCTVHAR: 43.1% -> 50.2%Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy