Generalized Category Discovery on Stanford Action, Location, Mood (Known)
96.2Action AccuracyCLIP-ZS
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| CLIP-ZSBackbone=ViT-B/162025.12 | 96.2 | 60.4 | 87.4 | 75 | |
| CLIP-ZS + GT vocabBackbone=ViT-B/162025.12 | 93.6 | 59.6 | 78.8 | 75 | |
| CLIP-ZS + LLM vocabBackbone=ViT-B/162025.12 | 93.5 | 58.3 | 75.9 | 75 | |
| GCDBackbone=ViT-B/162025.12 | 89.4 | 75.4 | 64.3 | 75 | |
| OAKBackbone=ViT-B/162025.12 | 88.9 | 83.9 | 68.8 | 0 | |
| SS-KMeansBackbone=ViT-B/162025.12 | 67 | 56.3 | 43.2 | 0 |