Zero-Shot Learning 벤치마크
Zero-Shot Learning on AwA2
average top-1 classification accuracy
- 2020-03-17 — ZSL_TF-VAEGAN: average top-1 classification accuracy 72.2
- 2020-06-18 — ZSL-KG: average top-1 classification accuracy 78.08
- 2024-06-05 — ZeroDiff: average top-1 classification accuracy 86.4
| Rank | Model | average top-1 classification accuracy | Accuracy Seen | Accuracy Unseen | H | Paper | Code | Year |
|---|---|---|---|---|---|---|---|---|
| 1 | ZeroDiff | 86.4 | – | – | – | Exploring Data Efficiency in Zero-Shot Learning with Diffusion Models | 2024 | |
| 2 | ZSL-KG | 78.08 | – | – | – | Zero-Shot Learning with Common Sense Knowledge Graphs | BatsResearch/zsl-kg · BatsResearch/nayak-arxiv20-code · batsresearch/nayak-tmlr22-code | 2020 |
| 3 | ZSL_TF-VAEGAN | 72.2 | – | – | – | Latent Embedding Feedback and Discriminative Features for Zero-Shot Classification | akshitac8/tfvaegan | 2020 |
| 4 | DUET (Ours) | 69.9 | 84.7 | 63.7 | 72.7 | DUET: Cross-modal Semantic Grounding for Contrastive Zero-shot Learning | zjukg/structure-clip · zjukg/DUET | 2022 |