Research map / Vision-language and multimodal / CLIP and zero-shot
Online Zero-Shot Classification with CLIP
Qi Qian, Juhua Hu
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 online label learning with alpha=1 over vanilla zero-shot (ablation) | OnZeta · ImageNet (ResNet-50) | accuracy (%): 60.32 -> 61.96 | Improved |
| add online vision proxy learning with pseudo-label mixing on top of online label learning (ablation) | OnZeta · ImageNet (ResNet-50) | accuracy (%): 61.96 -> 62.50 | Improved |
| use only the vision proxy at the end of the stream (beta=1) (ablation) | OnZeta · ImageNet (ResNet-50) | accuracy (%): 62.50 -> 61.84 | Hurt |
| use dynamic increasing combination ratio instead of fixed lambda (ablation) | OnZeta · ImageNet (ResNet-50) | accuracy (%): 62.28 -> 62.50 | Improved |
| visit data for additional epochs instead of a single online pass (ablation) | OnZeta · ImageNet (ResNet-50) | accuracy (%): 62.50 -> 63.46 | Improved |
| learn class proxy in vision space instead of optimizing text prompts (TPT) | OnZeta · ImageNet (ResNet-50) | accuracy (%): 60.74 -> 62.42 | Improved |
| online label learning only on 13 downstream tasks (ablation) | OnZeta · 13 downstream tasks avg (ResNet-50) | accuracy (%): 55.06 -> 57.84 | Improved |
| add vision proxy learning on 13 tasks (ablation) | OnZeta · 13 downstream tasks avg (ViT-B/16) | accuracy (%): 68.13 -> 68.69 | Improved |
| online method vs offline InMaP with whole unlabeled set | OnZeta · ImageNet (ResNet-50) | accuracy (%): 62.42 vs 63.74 | Hurt |
Closest papers by meaning
- Interpreting and Analysing CLIP's Zero-Shot Image Classification via Mutual Knowledge (NeurIPS 2024 · similarity 0.89)
- Enhancing Zero-Shot Vision Models by Label-Free Prompt Distribution Learning and Bias Correcting (NeurIPS 2024 · similarity 0.88)
- Transductive Zero-Shot and Few-Shot CLIP (CVPR 2024 · similarity 0.88)
- TPR: Topology-Preserving Reservoirs for Generalized Zero-Shot Learning (NeurIPS 2024 · similarity 0.87)
- OTTER: Effortless Label Distribution Adaptation of Zero-shot Models (NeurIPS 2024 · similarity 0.87)
- Data-Free Generalized Zero-Shot Learning (AAAI 2024 · 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.
