Generalized Category Discovery on Stanford Action, Location, Mood (Overall)
86.9Action AccuracyOAK
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| OAKBackbone=ViT-B/162025.12 | 86.9 | 85.9 | 78.4 | 70.3 | |
| CLIP-ZS + GT vocabBackbone=ViT-B/162025.12 | 86.7 | 59.7 | 72.1 | 38.3 | |
| GCDBackbone=ViT-B/162025.12 | 78.3 | 77.8 | 52.1 | 52.3 | |
| CLIP-ZS + LLM vocabBackbone=ViT-B/162025.12 | 65.2 | 47.5 | 55 | 43 | |
| SS-KMeansBackbone=ViT-B/162025.12 | 60.2 | 62.9 | 61.3 | 47.7 |