Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Feature Generating Networks for Zero-Shot Learning
Read the paperCheck 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 |
|---|---|---|---|
| Generate CNN features for unseen classes and train a classifier with them | ZSL classifiers · CUB, FLO, SUN, and AWA | Top-1 accuracy: 54.9% to 61.5%; 53.4% to 71.2%; 58.1% to 62.1%; 65.6% to 69.9% | Helped |
| Generate CNN features for unseen classes and train a classifier with them | GZSL classifiers · CUB, FLO, SUN, and AWA | H: 34.4% to 49.7%; 21.9% to 65.6%; 26.3% to 39.4%; 27.5% to 59.6% | Helped |
| Add a classification loss to the WGAN generator objective (ablation) | f-WGAN · CUB and FLO ZSL and GZSL | accuracy | Helped |
| Use feature generation rather than no generated data (ablation) | ZSL and GZSL classifiers · All evaluated datasets and classifiers | accuracy | Helped |
| Increase generated unseen-class features from 1 to 100 (ablation) | Softmax classifier · CUB and FLO GZSL | accuracy: 28.2% to 56.5%; 37.9% to 66.5% | Helped |
| Use f-CLSWGAN with GoogLeNet rather than ResNet-101 features (ablation) | f-CLSWGAN · CUB GZSL | H: 49.7% to 36.9% | Hurt |
| Use sentence rather than attribute class embeddings (ablation) | f-CLSWGAN · CUB GZSL | H: 49.7% to 54.0% | Helped |
| Use StackGAN-generated images rather than no generated data (ablation) | ALE · CUB GZSL | H: 45.1% to 31.9% | Hurt |
| Use StackGAN-generated images rather than no generated data (ablation) | ALE · FLO GZSL | H: 21.9% to 49.0% | Helped |
| Generate CNN features rather than StackGAN images (ablation) | ALE · CUB and FLO GZSL | H: 31.9% to 54.0%; 49.0% to 65.6% | Helped |
| Use f-CLSWGAN rather than f-GAN (ablation) | Feature generator · FLO held-out seen-class classification | accuracy | Helped |
| Use f-WGAN rather than f-GAN (ablation) | Feature generator · ZSL and GZSL | accuracy | Helped |
Closest papers by meaning
- Leveraging the Invariant Side of Generative Zero-Shot Learning (CVPR 2019 · similarity 0.93)
- Generative Dual Adversarial Network for Generalized Zero-Shot Learning (CVPR 2019 · similarity 0.92)
- A Generative Adversarial Approach for Zero-Shot Learning From Noisy Texts (CVPR 2018 · similarity 0.92)
- Gradient Matching Generative Networks for Zero-Shot Learning (CVPR 2019 · similarity 0.91)
- Data-Free Generalized Zero-Shot Learning (AAAI 2024 · 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)
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.