Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
ReGen: A good Generative Zero-Shot Video Classifier Should be Rewarded
Adrian Bulat, Enrique Sanchez, Brais Martinez, Georgios Tzimiropoulos
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 classification reward CLS-R with RL on top of REST (ablation) | REST GIT captioner · Kinetics-220 (1-vs-620) | Top-1 accuracy (%): 29.6 -> 38.6 | Improved |
| add CLS-R reward (ablation) | REST GIT captioner · VaTeX zero-shot captioning | CIDEr: 39.1 -> 28.1 | Hurt |
| add CLIP-R video-specific reward to CLS-R (ablation) | ReGen · HMDB-51 zero-shot | Top-1 accuracy (%): 55.0 -> 55.5 | Improved |
| add grammar reward to CLS-R and CLIP-R (ablation) | ReGen · VaTeX zero-shot captioning | CIDEr: 29.3 -> 40.5 | Improved |
| add grammar reward to CLS-R and CLIP-R (ablation) | ReGen · Kinetics-220 (1-vs-620) | Top-1 accuracy (%): 38.9 -> 38.2 | Mixed |
| add CLIP-R to CLS-R + grammar reward (ablation) | ReGen · VaTeX zero-shot captioning | CIDEr: 33.4 -> 40.5 | Improved |
| use discriminator-based grammar reward instead of LM-based (ablation) | ReGen · Kinetics-220 (1-vs-620) | Top-1 accuracy (%): 38.2 -> 35.1 accuracy; CIDEr 40.5 -> 46.1 | Mixed |
| use larger grammar LM (BERT-L vs BERT-B) (ablation) | ReGen · Kinetics-220 (1-vs-620) | Top-1 accuracy (%): 37.5 -> 38.2 | Improved |
| apply ReGen RL fine-tuning to BLIP architecture | REST BLIP · Kinetics-220 (1-vs-620) | Top-1 accuracy (%): 29.51 -> 37.6 | Improved |
Closest papers by meaning
- Distilling Vision-Language Models on Millions of Videos (CVPR 2024 · similarity 0.86)
- CapRL: Stimulating Dense Image Caption Capabilities via Reinforcement Learning (ICLR 2026 · similarity 0.86)
- Learning by Correction: Efficient Tuning Task for Zero-Shot Generative Vision-Language Reasoning (CVPR 2024 · similarity 0.86)
- VideoPoet: A Large Language Model for Zero-Shot Video Generation (ICML 2024 · similarity 0.85)
- OmniViD: A Generative Framework for Universal Video Understanding (CVPR 2024 · similarity 0.85)
- BalCapRL : A Balanced Framework for RL-Based MLLM Image Captioning (NeurIPS 2026 · 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.
