Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
CLASTER: Clustering with Reinforcement Learning for Zero-Shot Action Recognition
Shreyank N Gowda, Laura Sevilla-Lara, Frank Keller, Marcus Rohrbach
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 clustering of representation (ablation) | CLASTER ZSL · UCF101 ZSL | accuracy (%): 31.6 -> 46.4 | Improved |
| add clustering of representation (ablation) | CLASTER ZSL · HMDB51 ZSL | accuracy (%): 25.6 -> 36.8 | Improved |
| use meaningful (K-means) clusters instead of random clusters (ablation) | CLASTER ZSL · UCF101 ZSL | accuracy (%): 24.1 -> 35.3 | Improved |
| add random clustering (ablation) | CLASTER ZSL · UCF101 ZSL | accuracy (%): 31.6 -> 24.1 | Hurt |
| add semantic embedding in the clustered representation (ablation) | CLASTER ZSL · UCF101 ZSL | accuracy (%): 39.4 -> 46.4 | Improved |
| add semantic embedding in the clustered representation (ablation) | CLASTER ZSL · HMDB51 ZSL | accuracy (%): 27.5 -> 36.8 | Improved |
| optimize centroids with RL instead of SGD (ablation) | CLASTER ZSL · UCF101 ZSL | accuracy (%): 39.1 -> 46.4 | Improved |
| optimize centroids with RL instead of NetVLAD (ablation) | CLASTER ZSL · UCF101 ZSL | accuracy (%): 41.7 -> 46.4 | Improved |
| optimize clusters with RL instead of K-means (ablation) | CLASTER · UCF101 train purity | purity: 0.77 -> 0.89 | Improved |
| concatenate visual and semantic features instead of averaging (ablation) | CLASTER ZSL · HMDB51 ZSL (ED) | accuracy (%): 33.1 -> 43.2 | Improved |
| concatenate instead of dot product (ablation) | CLASTER ZSL · HMDB51 ZSL (ED) | accuracy (%): 33.9 -> 43.2 | Improved |
| combine word2vec and sen2vec embeddings | CLASTER · ZSL average over datasets | accuracy (%): 0.3%, 0.8% and 0.9% | Improved |
Closest papers by meaning
- I Know the Relationships: Zero-Shot Action Recognition via Two-Stream Graph Convolutional Networks and Knowledge Graphs (AAAI 2019 · similarity 0.87)
- Transductive Zero-Shot Action Recognition by Word-Vector Embedding (IJCV 2017 · similarity 0.87)
- Cross-Modal Representation Learning for Zero-Shot Action Recognition (CVPR 2022 · similarity 0.86)
- Rethinking Zero-Shot Learning: A Conditional Visual Classification Perspective (ICCV 2019 · similarity 0.86)
- Rethinking Zero-Shot Action Recognition: Learning from Latent Atomic Actions (ECCV 2022 · similarity 0.86)
- Control-oriented Clustering of Visual Latent Representation (ICLR 2025 · 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.
