Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Harnessing GANs for Zero-Shot Learning of New Classes in Visual Speech Recognition
Yaman Kumar, Dhruva Sahrawat, Shubham Maheshwari, Debanjan Mahata, Amanda Stent, Yifang Yin, Rajiv Ratn Shah, Roger Zimmermann
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 TC-GAN-generated videos for unseen English phrases | VSR classifier trained on Oulu-phrases · Oulu-phrases, 1–4 unseen classes | unseen-class top-3 accuracy: 0 -> 86.1; 0 -> 52.8; 0 -> 49.1; 0 -> 31.2 | Improved |
| Add TC-GAN-generated videos for unseen English phrases | VSR classifier trained on Oulu-phrases · Oulu-phrases, 1–4 unseen classes | unseen-class top-1 accuracy: 0 -> 33.3; 0 -> 23.6; 0 -> 21.3; 0 -> 16 | Improved |
| Add TC-GAN-generated videos for unseen classes | VSR classifier · Oulu-phrases, seen classes | top-3 accuracy: 96.9 -> 96; 96.5 -> 95.8; 98.8 -> 94.8; 98.1 -> 98.1 | No clear effect |
| Add TC-GAN-generated videos for unseen Hindi phrases | VSR classifier trained with English Oulu-phrases · Hindi phrases | unseen-class top-3 accuracy: 0 -> 100 | Improved |
| Add TC-GAN-generated videos for unseen Hindi phrases | VSR classifier trained with English Oulu-phrases · Hindi phrases | seen-class top-1 accuracy: 94.4 -> 94.7 | No clear effect |
| Replace TC-GAN-generated videos with viseme-concatenation videos | VSR classifier · Hindi unseen classes | unseen-class top-3 accuracy: 100 -> 32.4 | Hurt |
| Use generated videos with inter-visemic frames instead of discontinuous concatenated viseme frames | VSR classifier · Oulu-phrases, unseen classes | unseen-class accuracy | Improved |
| Train with TC-GAN-generated videos for all classes when no real phrase classes are available | VSR classifier · Oulu-phrases cold-start (10 unseen classes) | overall top-3 accuracy: 30.8 -> 70 | Improved |
| Train with TC-GAN-generated videos for all classes when no real phrase classes are available | VSR classifier · Oulu-phrases cold-start (10 unseen classes) | overall top-1 accuracy: 13.3 -> 40 | Improved |
Closest papers by meaning
- Feature Generating Networks for Zero-Shot Learning (CVPR 2018 · similarity 0.88)
- A Generative Adversarial Approach for Zero-Shot Learning From Noisy Texts (CVPR 2018 · similarity 0.88)
- Zero-AVSR: Zero-Shot Audio-Visual Speech Recognition with LLMs by Learning Language-Agnostic Speech Representations (ICCV 2025 · similarity 0.87)
- On the Zero-shot Adversarial Robustness of Vision-Language Models: A Truly Zero-shot and Training-free Approach (CVPR 2025 · similarity 0.87)
- Audio-visual Generalized Zero-shot Learning the Easy Way (ECCV 2024 · similarity 0.87)
- VSRo-200: A Romanian Visual Speech Recognition Dataset for Studying Supervision and Multimodal Robustness (NeurIPS 2026 · 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.
