Label Propagation for Zero-shot Classification with Vision-Language Models

CVPR 2024  ยท  Vladan Stojniฤ‡, Yannis Kalantidis, Giorgos Tolias ยท

Vision-Language Models (VLMs) have demonstrated impressive performance on zero-shot classification, i.e. classification when provided merely with a list of class names. In this paper, we tackle the case of zero-shot classification in the presence of unlabeled data. We leverage the graph structure of the unlabeled data and introduce ZLaP, a method based on label propagation (LP) that utilizes geodesic distances for classification. We tailor LP to graphs containing both text and image features and further propose an efficient method for performing inductive inference based on a dual solution and a sparsification step. We perform extensive experiments to evaluate the effectiveness of our method on 14 common datasets and show that ZLaP outperforms the latest related works. Code: https://github.com/vladan-stojnic/ZLaP

PDF Abstract CVPR 2024 PDF CVPR 2024 Abstract

Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Zero-Shot Learning Caltech-101 ZLaP Accuracy 84 # 1
Zero-Shot Learning Caltech-101 ZLaP* Accuracy 83.1 # 2
Transductive Zero-Shot Classification Caltech-101 ZLaP Accuracy 83.7 # 1
Zero-Shot Learning CIFAR-10 ZLaP Accuracy 93.4 # 2
Transductive Zero-Shot Classification CIFAR-10 ZLaP Accuracy 93.6 # 1
Zero-Shot Learning CIFAR-10 ZLaP* Accuracy 93.6 # 1
Zero-Shot Learning CIFAR-100 ZLaP Accuracy 74 # 2
Zero-Shot Learning CIFAR-100 ZLaP* Accuracy 74.2 # 1
Transductive Zero-Shot Classification CIFAR-100 ZLaP Accuarcy 73.3 # 1
Transductive Zero-Shot Classification CUB-200-2011 ZLaP Accuracy 64.1 # 1
Zero-Shot Learning CUB-200-2011 ZLaP* Accuracy 64.2 # 2
Zero-Shot Learning CUB-200-2011 ZLaP Accuracy 64.3 # 1
Zero-Shot Learning DTD ZLaP* Accuracy 51 # 2
Zero-Shot Learning DTD ZLaP Accuracy 51.2 # 1
Transductive Zero-Shot Classification DTD ZLaP Accuracy 51.8 # 1
Zero-Shot Learning EuroSAT ZLaP Accuracy 63.2 # 1
Zero-Shot Learning EuroSAT ZLaP* Accuracy 62.7 # 2
Transductive Zero-Shot Classification EuroSAT ZLaP Accuracy 60.9 # 1
Zero-Shot Learning FGVC-Aircraft ZLaP Accuracy 29.1 # 1
Transductive Zero-Shot Classification FGVC-Aircraft ZLaP Accuracy 28.4 # 1
Zero-Shot Learning FGVC-Aircraft ZLaP* Accuracy 29 # 2
Transductive Zero-Shot Classification Flowers-102 ZLaP Accuracy 73.4 # 1
Zero-Shot Learning Flowers-102 ZLaP* Accuracy 75.5 # 2
Zero-Shot Learning Flowers-102 ZLaP Accuracy 75.9 # 1
Transductive Zero-Shot Classification Food-101 ZLaP Accuracy 87.9 # 1
Zero-Shot Learning Food-101 ZLaP* Accuracy 87.9 # 1
Zero-Shot Learning Food-101 ZLaP Accuracy 87.8 # 2
Zero-Shot Learning ImageNet ZLaP Top 1 Accuracy 72.1 # 1
Zero-Shot Learning ImageNet ZLaP* Top 1 Accuracy 72.1 # 1
Transductive Zero-Shot Classification ImageNet ZLaP Top 1 Accuracy 72.7 # 1
Zero-Shot Learning Oxford-IIIT Pets ZLaP* Accuracy 89 # 2
Zero-Shot Learning Oxford-IIIT Pets ZLaP Accuracy 90 # 1
Transductive Zero-Shot Classification Oxford-IIIT Pets ZLaP Accuracy 92.8 # 1
Zero-Shot Learning Stanford Cars ZLaP* Accuracy 71.8 # 1
Transductive Zero-Shot Classification Stanford Cars ZLaP Accuracy 72.1 # 1
Zero-Shot Learning Stanford Cars ZLaP Accuracy 71.2 # 2
Zero-Shot Learning SUN397 ZLaP Accuracy 71 # 2
Transductive Zero-Shot Classification SUN397 ZLaP Accuracy 71.9 # 1
Zero-Shot Learning SUN397 ZLaP* Accuracy 71.4 # 1
Zero-Shot Learning UCF101 ZLaP Accuracy 76.3 # 1
Transductive Zero-Shot Classification UCF101 ZLaP Accuracy 77.7 # 1
Zero-Shot Learning UCF101 ZLaP* Accuracy 76.3 # 1

Methods


ZLaP โ€ข ZLaP*