Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Zero-Shot Visual Recognition Using Semantics-Preserving Adversarial Embedding Networks
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 |
|---|---|---|---|
| Use independent reconstruction embedding F and adversarial transfer between F and E, rather than classification alone (ablation) | SP-AEN · SUN generalized ZSL | H (%): 21.6 -> 30.3 | Helped |
| Use independent reconstruction embedding F and adversarial transfer between F and E, rather than classification alone (ablation) | SP-AEN · CUB generalized ZSL | H (%): 32.9 -> 46.6 | Helped |
| Use independent reconstruction embedding F and adversarial transfer between F and E, rather than classification alone (ablation) | SP-AEN · AWA generalized ZSL | H (%): 28.5 -> 37.1 | Helped |
| Use independent reconstruction embedding F and adversarial transfer between F and E, rather than classification alone (ablation) | SP-AEN · aPY generalized ZSL | H (%): 10.2 -> 22.6 | Helped |
| Jointly train discriminative E and reconstructive G as an SAE-style autoencoder (ablation) | SP-AEN reconstruction · CUB, SUN, AWA and aPY unseen test images | mean squared pixel-level loss: SUN: 0.053 -> 0.285; CUB: 0.040 -> 0.281; AWA: 0.047 -> 0.259; aPY: 0.055 -> 0.275 | Hurt |
| Split the encoder output into classification and reconstruction branches (ablation) | SP-AEN reconstruction · CUB, SUN, AWA and aPY unseen test images | mean squared pixel-level loss: SUN: 0.053 -> 0.070; CUB: 0.040 -> 0.058; AWA: 0.047 -> 0.059; aPY: 0.055 -> 0.076 | Mixed |
| Use only the discriminative embedding E for reconstruction instead of an independent reconstructive embedding (ablation) | SP-AEN reconstruction · AWA and aPY unseen test images | mean squared pixel-level loss: AWA: 0.047 -> 0.075; aPY: 0.055 -> 0.085 | Hurt |
| Adversarially transfer semantics between reconstruction and classification embeddings (ablation) | SP-AEN · SUN, CUB, AWA and aPY generalized ZSL | H (%): over 10% on all datasets | Helped |
| Use the full objective instead of classification loss alone (ablation) | SP-AEN · SUN, CUB, AWA and aPY generalized ZSL | AUSUC | Helped |
| Use SP-AEN instead of the strongest reported baseline | zero-shot visual recognition · SUN, CUB, AWA and aPY | H (%): 12.2%, 9.3%, 4.0%, and 3.6% absolute performance difference | Helped |
Closest papers by meaning
- Zero-Shot Learning from Adversarial Feature Residual to Compact Visual Feature (AAAI 2020 · similarity 0.90)
- Generalized Zero-Shot Recognition Based on Visually Semantic Embedding (CVPR 2019 · similarity 0.90)
- Generative Dual Adversarial Network for Generalized Zero-Shot Learning (CVPR 2019 · similarity 0.90)
- Dual Adversarial Semantics-Consistent Network for Generalized Zero-Shot Learning (NeurIPS 2019 · similarity 0.90)
- Progressive Semantic-Guided Vision Transformer for Zero-Shot Learning (CVPR 2024 · similarity 0.89)
- ZeroMamba: Exploring Visual State Space Model for Zero-Shot Learning (AAAI 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.