Research map / Image generation / Editing and style transfer
A Face-to-Face Neural Conversation Model
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use face gestures alongside text instead of text alone | Text conversation model · MovieChat test, text | F1 (beam=1): 17.12 -> 18.69 | Helped |
| Use real face gestures rather than randomized face input (ablation) | Text+Face · MovieChat test, text | F1 (beam=1): 17.27 -> 18.69 | Helped |
| Add text information to gesture generation | Face gesture model · MovieChat test, gesture | F1 (beam=1): 12.96 -> 14.41 | Helped |
| Use real text rather than randomized text input (ablation) | Face+Text · MovieChat test, gesture | F1 (beam=1): 13.15 -> 14.41 | Helped |
| Use the History-FC history connection instead of History-RNN (ablation) | History model · MovieChat test, text | F1 (beam=1): 19.59 -> 19.88 | Helped |
| Use the History-FC history connection instead of History-RNN (ablation) | History model · MovieChat test, gesture | F1 (beam=1): 9.81 -> 9.83 | No clear effect |
| Optimize with F1 policy-gradient reward instead of MLE (ablation) | Face-to-Face model · MovieChat test, text | F1 (beam=1): 20.17 -> 20.42 | Helped |
| Use adversarial reward instead of MLE (ablation) | Face-to-Face model · MovieChat test, text | F1 (beam=1): 20.17 -> 20.44 | Helped |
| Use adversarial reward instead of direct F1 reward (ablation) | Face-to-Face model · MovieChat test, text | F1 (beam=1): 20.42 -> 20.44 | No clear effect |
| Optimize with F1 policy-gradient reward instead of MLE (ablation) | Face-to-Face model · MovieChat test, gesture | F1 (beam=1): 18.07 -> 18.28 | Helped |
| Add the micro-gesture RNN (ablation) | NeuralHank decoder · NeuralHank user study, pairwise vs noMicro-noGAN | face preference (%): 39.0 -> 61.0 | Helped |
| Add adversarial training to the micro-gesture model (ablation) | NeuralHank decoder · NeuralHank user study, pairwise vs Micro-noGAN | overall preference (%): 46.7 -> 53.3 | Helped |
Closest papers by meaning
- A Knowledge-Grounded Neural Conversation Model (AAAI 2018 · similarity 0.87)
- Visual Dialog (CVPR 2017 · similarity 0.87)
- Infusing Multi-Source Knowledge with Heterogeneous Graph Neural Network for Emotional Conversation Generation (AAAI 2021 · similarity 0.87)
- Building End-To-End Dialogue Systems Using Generative Hierarchical Neural Network Models (AAAI 2016 · similarity 0.87)
- Best of Both Worlds: Transferring Knowledge from Discriminative Learning to a Generative Visual Dialog Model (NeurIPS 2017 · similarity 0.86)
- Faces that Speak: Jointly Synthesising Talking Face and Speech from Text (CVPR 2024 · similarity 0.85)
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.