Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Latent Embedding Feedback and Discriminative Features for Zero-Shot Classification
Sanath Narayan, Akshita Gupta, Fahad Shahbaz Khan, Cees G. M. Snoek, Ling Shao
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 feedback module (ablation) | TF-VAEGAN baseline · CUB ZSL (inductive) | top-1 accuracy: 61.2 -> 62.8 | Improved |
| add feedback module (ablation) | TF-VAEGAN baseline · CUB GZSL (inductive) | harmonic mean H: 53.5 -> 54.8 | Improved |
| add discriminative feature transformation (T-feature) (ablation) | TF-VAEGAN baseline · CUB ZSL (inductive) | top-1 accuracy: 61.2 -> 64.0 | Improved |
| add discriminative feature transformation (T-feature) (ablation) | TF-VAEGAN baseline · CUB GZSL (inductive) | harmonic mean H: 53.5 -> 56.9 | Improved |
| combine feedback and T-feature (ablation) | f-VAEGAN baseline · CUB GZSL (inductive) | harmonic mean H: 53.5 -> 58.1 | Improved |
| combine feedback and T-feature (ablation) | f-VAEGAN baseline · CUB ZSL (transductive) | top-1 accuracy: 70.6 -> 74.7 | Improved |
| combine feedback and T-feature (ablation) | f-VAEGAN baseline · CUB GZSL (transductive) | harmonic mean H: 63.7 -> 71.0 | Improved |
| take feedback input from semantic embedding decoder instead of discriminator (two-stage) (ablation) | feedback module · CUB ZSL (inductive) | top-1 accuracy: 61.4 -> 62.0 | Improved |
| train feedback with alternating strategy instead of two-stage (ablation) | feedback module · CUB GZSL (inductive) | harmonic mean H: 53.8 -> 54.8 | Improved |
| concatenate reconstructed attributes (ConcatFeat) to features (ablation) | f-VAEGAN baseline · CUB ZSL (inductive) | top-1 accuracy: +1.6% | Improved |
| concatenate latent embeddings h (T-feature) to features (ablation) | f-VAEGAN baseline · CUB GZSL (inductive) | harmonic mean H: +3.4% | Improved |
| integrate decoder and feedback into f-CLSWGAN (TF-CLSWGAN) | f-CLSWGAN · four datasets ZSL/GZSL | accuracy | Improved |
Closest papers by meaning
- Generative Dual Adversarial Network for Generalized Zero-Shot Learning (CVPR 2019 · similarity 0.91)
- Adaptive and Generative Zero-Shot Learning (ICLR 2021 · similarity 0.91)
- Generative Zero-Shot Learning via Low-Rank Embedded Semantic Dictionary (TPAMI 2018 · similarity 0.91)
- Zero-Shot Learning via Class-Conditioned Deep Generative Models (AAAI 2018 · similarity 0.91)
- Feature Generating Networks for Zero-Shot Learning (CVPR 2018 · similarity 0.91)
- Leveraging Seen and Unseen Semantic Relationships for Generative Zero-Shot Learning (ECCV 2020 · similarity 0.90)
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.
