Image Classification on Five Datasets 8-shot
76.8AccuracyFully Supervised w/ RA
Evaluation results
| Method | Links | |
|---|---|---|
| Fully Supervised w/ RALearning Paradigm=FT, Backbone=OpenCLIP ViT-B/32, Retrieval Augmentation=true2025.12 | 76.8 | |
| SWIFTLearning Paradigm=FT, Backbone=OpenCLIP ViT-B/32, Retrieval Augmentation=true2025.12 | 76.3 | |
| Fully SupervisedLearning Paradigm=FT, Backbone=OpenCLIP ViT-B/322025.12 | 75.4 | |
| SWATLearning Paradigm=FT, Backbone=OpenCLIP ViT-B/32, Retrieval Augmentation=true2025.12 | 71 | |
| DebiasPL on VLM (ours)Learning Paradigm=FT, Backbone=OpenCLIP ViT-B/322025.12 | 67.6 | |
| FS-FTLearning Paradigm=FT, Backbone=OpenCLIP ViT-B/322025.12 | 65.8 | |
| FixMatch on VLM (ours)Learning Paradigm=FT, Backbone=OpenCLIP ViT-B/322025.12 | 65.7 | |
| FineSSLLearning Paradigm=PL, Backbone=OpenCLIP ViT-B/322025.12 | 64.6 | |
| CLAPLearning Paradigm=AL, Backbone=OpenCLIP ViT-B/322025.12 | 61.3 | |
| Linear ProbingLearning Paradigm=AL, Backbone=OpenCLIP ViT-B/322025.12 | 61.2 | |
| CMLPLearning Paradigm=AL, Backbone=OpenCLIP ViT-B/322025.12 | 59.4 | |
| TaskRes(e)Learning Paradigm=AL, Backbone=OpenCLIP ViT-B/322025.12 | 58 | |
| CLIP-AdapterLearning Paradigm=AL, Backbone=OpenCLIP ViT-B/322025.12 | 55.2 | |
| PLOTLearning Paradigm=PL, Backbone=OpenCLIP ViT-B/322025.12 | 54.9 | |
| CoOpLearning Paradigm=PL, Backbone=OpenCLIP ViT-B/322025.12 | 54.6 | |
| Tip-Adapter (f)Learning Paradigm=AL, Backbone=OpenCLIP ViT-B/322025.12 | 52.9 | |
| FixMatch on VLMLearning Paradigm=FT, Backbone=OpenCLIP ViT-B/322025.12 | 49.9 | |
| DebiasPL on VLMLearning Paradigm=FT, Backbone=OpenCLIP ViT-B/322025.12 | 49.8 |