Research map / Vision-language and multimodal / CLIP and zero-shot
Prompt Tuning of Deep Neural Networks for Speaker-Adaptive Visual Speech Recognition
Minsu Kim, Hyung-Il Kim, Yong Man Ro
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 |
|---|---|---|---|
| tune speaker prompts on 5 min of adaptation data | pre-trained VSR · GRID | mean WER: 12.04 -> 3.80 | Improved |
| tune speaker prompts on 5 min of adaptation data | pre-trained VSR · LRW-ID | mean accuracy: 87.54 -> 89.99 | Improved |
| add concatenation prompt to Add and Pad prompts (ablation) | CNN prompts · GRID | mean WER (5 min): 4.99 -> 4.04 (Pad vs Add+Pad); 4.17 -> 3.80 (Add vs Add+Pad+Cat) | Improved |
| use padding prompt instead of addition prompt | single prompt, deep CNN · LRW-ID | mean accuracy (5 min): 88.03 -> 89.62 | Improved |
| use addition prompt instead of padding prompt | single prompt, shallow CNN · GRID | mean WER (5 min): 4.99 -> 4.17 | Improved |
| increase adaptation data from 1 to 5 min | Add+Pad+Cat · GRID | mean WER: 5.53 -> 3.80 | Improved |
| add padding-prompt layers (N_l 5 -> 17) (ablation) | padding prompt · LRW-ID | accuracy (5 min): 88.55 -> 89.62 | Improved |
| increase concatenation prompt length N_p 1 -> 5 (ablation) | concatenation prompt · LRW-ID | accuracy (1 and 3 min): 88.61 -> 88.92 (3 min); 88.27 -> 88.28 (1 min) | Improved |
| use prompt tuning instead of full finetuning (small data) | Pad+Cat prompt · LRW-ID | accuracy (5 min): 89.91 -> 89.99 | Improved |
| use prompt tuning instead of full finetuning | Add+Pad+Cat prompt · GRID | WER (5 min): 3.60 -> 3.80 | Hurt |
| use prompt tuning instead of full finetuning with 100% adaptation data | Add+Pad+Cat prompt · GRID | WER: 2.34 -> 3.13 | Hurt |
| use prompt tuning instead of Speaker code | Add+Cat prompt · GRID | WER (5 min): 4.68 -> 4.00 | Improved |
Closest papers by meaning
- Learning to Learn Better Visual Prompts (AAAI 2024 · similarity 0.88)
- Black-Box Test-Time Prompt Tuning for Vision-Language Models (AAAI 2025 · similarity 0.87)
- Language Models as Black-Box Optimizers for Vision-Language Models (CVPR 2024 · similarity 0.87)
- LaViP: Language-Grounded Visual Prompting (AAAI 2024 · similarity 0.86)
- Skip Tuning: Pre-trained Vision-Language Models are Effective and Efficient Adapters Themselves (CVPR 2025 · similarity 0.86)
- Amend to Alignment: Decoupled Prompt Tuning for Mitigating Spurious Correlation in Vision-Language Models (ICML 2024 · 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.
