Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
An Empirical Study and Analysis of Generalized Zero-Shot Learning for Object Recognition in the Wild
Wei-Lun Chao, Soravit Changpinyo, Boqing Gong, Fei Sha
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 |
|---|---|---|---|
| use calibrated stacking instead of Gaussian novelty detection | SynC struct · AwA | AUSUC: 0.424 -> 0.583 | Improved |
| use calibrated stacking instead of LoOP novelty detection | SynC struct · CUB | AUSUC: 0.224 -> 0.356 | Improved |
| use calibrated stacking instead of Gaussian novelty detection | SynC struct · SUN | AUSUC: 0.151 -> 0.260 | Improved |
| calibrate seen-class scores (gamma) instead of direct stacking | SynC struct · AwA | unseen-to-all accuracy (%): 0.4 -> 73.4 | Improved |
| select hyperparameters by AUSUC instead of accuracies (ablation) | SynC o-vs-o · AwA | AUSUC: 0.539 -> 0.568 | Improved |
| select hyperparameters by AUSUC instead of accuracies (ablation) | ConSE · AwA | AUSUC: 0.443 -> 0.428 | Hurt |
| select hyperparameters by AUSUC instead of accuracies (ablation) | DAP · CUB | AUSUC: 0.202 -> 0.194 | Hurt |
| select hyperparameters by AUSUC instead of accuracies (ablation) | SynC struct · CUB | AUSUC: 0.356 -> 0.356 | No clear effect |
| replace word2vec embeddings with G-attr from all images (ablation) | SynC o-vs-o · ImageNet-2K | AUSUC, Flat hit@1: 0.04 -> 0.25 | Improved |
| increase labeled images for G-attr from 1 to 100 (ablation) | SynC o-vs-o · ImageNet-2K | AUSUC, Flat hit@1: 0.08 -> 0.25 | Improved |
| replace human visual attributes with G-attr from all images (ablation) | SynC o-vs-o · CUB | AUSUC: 0.34 -> 0.54 | Improved |
| replace visual attributes with 1-shot G-attr (ablation) | SynC o-vs-o · AwA | AUSUC: 0.57 -> 0.55 | Hurt |
Closest papers by meaning
- A Review of Generalized Zero-Shot Learning Methods (TPAMI 2022 · similarity 0.94)
- Improving Generalized Zero-Shot Learning by Exploring the Diverse Semantics from External Class Names (CVPR 2024 · similarity 0.93)
- On Zero-Shot Recognition of Generic Objects (CVPR 2019 · similarity 0.93)
- Generalized Zero-Shot Recognition Based on Visually Semantic Embedding (CVPR 2019 · similarity 0.92)
- Data Distribution Distilled Generative Model for Generalized Zero-Shot Recognition (AAAI 2024 · similarity 0.91)
- Data-Free Generalized Zero-Shot Learning (AAAI 2024 · similarity 0.91)
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.
