Research map / 3D vision / Face and shape
EmoteGPT: 3D Human Facial Expression from Natural Language Descriptions
Haoran Wang, Mohit Mendiratta, Christian Theobalt, Adam Kortylewski
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 VQA instruction data to text-only training (ablation) | EmoteGPT · Txt2Emote explicit | emotion accuracy: 0.52 -> 0.56 | Improved |
| add face-image expression data on top of text+VQA (ablation) | EmoteGPT · Txt2Emote explicit | emotion accuracy: 0.56 -> 0.59 | Improved |
| add VQA and image data (ablation) | EmoteGPT · Txt2Emote implicit | arousal CCC: 0.37 -> 0.48 | Improved |
| add VQA data alone (ablation) | EmoteGPT · Txt2Emote implicit | valence CCC: 0.55 -> 0.55 | No clear effect |
| use dedicated <Expr> token instead of mean pooling (ablation) | EmoteGPT · Txt2Emote implicit | L1 face: 1.15 -> 1.09 | Improved |
| use <Expr> token instead of <EOS> token (ablation) | EmoteGPT · Txt2Emote implicit | L1 face: 1.43 -> 1.09 | Improved |
| combine FLAME parameter loss with weighted mesh geometry loss (ablation) | EmoteGPT · Txt2Emote implicit | L1 face: 1.14 (FLAME only) / 1.13 (geo only) -> 1.09 | Improved |
| use MLLM (LLaVA-1.5) instead of CLIP regressor | EmoteGPT · Txt2Emote implicit | valence CCC: 0.49 -> 0.57 | Improved |
| train without implicit descriptions, test on implicit | EmoteGPT · Txt2Emote implicit | valence CCC: CLIP 0.36 vs EmoteGPT 0.49 | Improved |
| regress 3DMM parameters from text vs diffusion text-to-3D head | EmoteGPT · Txt2Emote explicit (50 samples) | L1 face: 1.51 (Portrait3D) -> 1.01 | Improved |
| direct text-to-3DMM vs text-to-image then fit FLAME | EmoteGPT · Txt2Emote implicit | L1 face: 2.12 (Z-Image-Turbo) -> 1.09 | Improved |
| CLIP baseline vs EmoteGPT on explicit arousal CCC (explicit-only training) | EmoteGPT · Txt2Emote explicit | arousal CCC: CLIP 0.57 vs EmoteGPT 0.50 | Hurt |
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.
- HumanNorm — HumanNorm: Learning Normal Diffusion Model for High-quality and Realistic 3D Human Generation (CVPR 2024) · 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
- EMOPortraits: Emotion-enhanced Multimodal One-shot Head Avatars (CVPR 2024 · similarity 0.89)
- DEEPTalk: Dynamic Emotion Embedding for Probabilistic Speech-Driven 3D Face Animation (AAAI 2025 · similarity 0.89)
- ExpCLIP: Bridging Text and Facial Expressions via Semantic Alignment (AAAI 2024 · similarity 0.88)
- TEASER: Token Enhanced Spatial Modeling for Expressions Reconstruction (ICLR 2025 · similarity 0.88)
- High-Fidelity 3D Head Avatars Reconstruction through Spatially-Varying Expression Conditioned Neural Radiance Field (AAAI 2024 · similarity 0.88)
- 3D Facial Expressions through Analysis-by-Neural-Synthesis (CVPR 2024 · similarity 0.88)
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.
