Research map / Detection and segmentation / Open-set and few-shot detection
Synthesized Classifiers for Zero-Shot Learning
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Synthesize classifiers from phantom-class bases rather than use ConSE | Zero-shot classification on ImageNet · ImageNet All | Flat hit@1: 1.4 -> 1.5 | Mixed |
| Use one-versus-other loss instead of structured Crammer-Singer loss (ablation) | Synthesized classifiers · AwA | accuracy (%): 72.9 -> 69.7 | Hurt |
| Use structured Crammer-Singer loss instead of one-versus-other loss (ablation) | Synthesized classifiers · CUB | accuracy (%): 53.4 -> 54.7 | Helped |
| Use structured Crammer-Singer loss instead of one-versus-other loss (ablation) | Synthesized classifiers · SUN | accuracy (%): 62.8 -> 62.7 | No clear effect |
| Use 1000-dimensional rather than 100-dimensional word vectors (ablation) | Ours o-vs-o on AwA · AwA | accuracy (%): 42.2 -> 57.5 | Helped |
| Combine attributes with 1000-dimensional word vectors (ablation) | Ours o-vs-o on AwA · AwA | accuracy (%): 69.7 -> 76.3 | Helped |
| Learn phantom semantic embeddings (ablation) | Ours o-vs-o on AwA · AwA, attributes | accuracy (%): 69.7% -> 71.1% | Helped |
| Learn phantom semantic embeddings (ablation) | Ours o-vs-o on AwA · AwA, 1000-d word vectors | accuracy (%): 57.6% -> 56.6% | Hurt |
| Learn phantom semantic embeddings (ablation) | Ours o-vs-o on CUB · CUB | accuracy (%): 53.4% -> 54.2% | Helped |
| Tune hyperparameters with class-wise rather than sample-wise cross-validation (ablation) | Ours o-vs-o on CUB · CUB, GoogLeNet | accuracy (%): 52.0 -> 53.4 | Helped |
| Learn the semantic similarity metric (ablation) | Ours o-vs-o on AwA · AwA, attributes | accuracy (%): 69.7% -> 73.4% | Helped |
| Increase the number of seen classes (ablation) | Ours on CUB with 50 unseen classes · CUB | accuracy: 38.4 -> 49.9 -> 53.8 | Helped |
Closest papers by meaning
- Zero-Shot Recognition Using Dual Visual-Semantic Mapping Paths (CVPR 2017 · similarity 0.91)
- Semi-Supervised Vocabulary-Informed Learning (CVPR 2016 · similarity 0.91)
- Adversarial Zero-shot Learning With Semantic Augmentation (AAAI 2018 · similarity 0.90)
- Context-Aware Zero-Shot Learning for Object Recognition (ICML 2019 · similarity 0.90)
- TPR: Topology-Preserving Reservoirs for Generalized Zero-Shot Learning (NeurIPS 2024 · similarity 0.90)
- Class Distribution Shifts in Zero-Shot Learning: Learning Robust Representations (NeurIPS 2024 · similarity 0.88)
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.