Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Zero-Shot Learning Using Synthesised Unseen Visual Data with Diffusion Regularisation
Yang Long, Li Liu, Fumin Shen, Ling Shao, Xuelong Li
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 graph regularisation to linear regression (ablation) | linear regression synthesis · CUB unseen (CA, NN) | accuracy: 27.28 -> 38.82 | Improved |
| add diffusion regularisation to linear regression (ablation) | linear regression synthesis · aPY unseen (CA, NN) | accuracy: 17.96 -> 30.28 | Improved |
| add diffusion regularisation to graph-only (ablation) | GR-only · CUB unseen (CA, NN) | accuracy: 38.82 -> 44.90 | Improved |
| add graph regularisation to diffusion-only (ablation) | DR-only · aPY unseen (CA, NN) | accuracy: 30.28 -> 42.25 | Improved |
| add both regularisations to linear regression (ablation) | linear regression synthesis · SUN unseen (CA, NN) | accuracy: 63.00 -> 80.50 | Improved |
| use SVM on synthesised samples instead of NN | Ours sample-based · aPY unseen | accuracy: 38.49 -> 53.21 | Improved |
| use SVM on synthesised samples instead of NN | Ours sample-based · CUB unseen | accuracy: 39.82 -> 45.72 | Improved |
| use mean of synthesised features instead of class-level attribute average | Ours prototype · CUB / SUN / aPY unseen | accuracy: CUB 44.90 -> 44.90; SUN 80.50 -> 78.50; aPY 42.25 -> 41.97 | No clear effect |
| remove diffusion regularisation: variance of synthesised features (ablation) | UVDS · SUN/aPY/CUB synthesised data | variance balance | Improved |
| use GAN to synthesise unseen features instead of regression | GAN · AwA (VGG-19, CA, PC) | per-class accuracy: 82.12 -> 62.40 | Hurt |
| use UVDS instead of baselines in GZSL on CUB | SynC · CUB GZSL U-T | accuracy: 13.2 -> 23.8 | Improved |
| use UVDS on large-scale ImageNet 2-hops | SYNC · ImageNet 2H | top-1 accuracy: 9.26 -> 10.15 | Improved |
Closest papers by meaning
- ZeroDiff: Solidified Visual-semantic Correlation in Zero-Shot Learning (ICLR 2025 · similarity 0.92)
- From Zero-Shot Learning to Conventional Supervised Classification: Unseen Visual Data Synthesis (CVPR 2017 · similarity 0.92)
- ZeroDiff++: Generative Test-Time Adaptation for Zero-shot Learning (TPAMI 2026 · similarity 0.92)
- Rethinking Zero-Shot Learning: A Conditional Visual Classification Perspective (ICCV 2019 · similarity 0.91)
- Generalized Zero-Shot Recognition Based on Visually Semantic Embedding (CVPR 2019 · similarity 0.91)
- Progressive Semantic-Guided Vision Transformer for Zero-Shot Learning (CVPR 2024 · 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.
