Research map / Video / Video generation
Memories are One-to-Many Mapping Alleviators in Talking Face Generation
Anni Tang, Tianyu He, Xu Tan, Jun Ling, Runnan Li, Sheng Zhao, Jiang Bian, Li Song
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add implicit memory to the audio-to-expression model (ablation) | audio-to-expression model · GRID | RMSE: 0.0834 -> 0.0826 | Improved |
| add implicit memory to the audio-to-expression model (ablation) | audio-to-expression model · 15s HDTF adaptation | RMSE: 0.1448 -> 0.1343 | Improved |
| use implicit rather than explicit memory in audio-to-expression (ablation) | audio-to-expression model · GRID | RMSE: 0.0846 -> 0.0826 | Improved |
| add explicit memory to the neural-rendering model (ablation) | neural-rendering model · Obama | LPIPS: 0.0248 -> 0.0207 | Improved |
| add explicit memory to the neural-rendering model (ablation) | neural-rendering model · 15s HDTF adaptation | LPIPS: 0.0254 -> 0.0218 | Improved |
| use explicit rather than implicit memory in neural rendering (ablation) | neural-rendering model · 15s HDTF adaptation | LPIPS: 0.0264 -> 0.0218 | Improved |
| increase implicit memory from 500 to 1000 key-value pairs (ablation) | audio-to-expression model · GRID | RMSE: 0.0828 -> 0.0826 | Improved |
| increase implicit memory from 1000 to 2000 key-value pairs (ablation) | audio-to-expression model · GRID | RMSE: 0.0826 -> 0.0825 | No clear effect |
| increase explicit memory from 100 to 300 key-value pairs (ablation) | neural-rendering model · Obama | LPIPS: 0.0218 -> 0.0207 | Improved |
| increase explicit memory from 300 to 500 key-value pairs (ablation) | neural-rendering model · Obama | LPIPS: 0.0207 -> 0.0205 | No clear effect |
| increase memory dimension from 64 to 128 (ablation) | implicit memory · GRID vs 15s HDTF adaptation | RMSE: 0.0826 -> 0.0823 on GRID; 0.1343 -> 0.1357 on HDTF | Mixed |
| use MemFace instead of ADNeRF | talking face generation · Obama | MOS overall: 3.700 -> 4.318 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- LipSync3D — LipSync3D: Data-Efficient Learning of Personalized 3D Talking Faces From Video Using Pose and Lighting Normalization (CVPR 2021) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Implicit Identity Representation Conditioned Memory Compensation Network for Talking Head video Generation (ICCV 2023 · similarity 0.88)
- SyncTalkFace: Talking Face Generation with Precise Lip-Syncing via Audio-Lip Memory (AAAI 2022 · similarity 0.88)
- From Inference to Generation: End-to-end Fully Self-supervised Generation of Human Face from Speech (ICLR 2020 · similarity 0.88)
- From Faces to Voices: Learning Hierarchical Representations for High-quality Video-to-Speech (CVPR 2025 · similarity 0.87)
- Long-Term TalkingFace Generation via Motion-Prior Conditional Diffusion Model (ICML 2025 · similarity 0.87)
- EmotiveTalk: Expressive Talking Head Generation through Audio Information Decoupling and Emotional Video Diffusion (CVPR 2025 · similarity 0.86)
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.
