Vision-Language Factuality Control on Flickr8k (test)
97.27ECRCFC-P-F
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| CFC-P-FBase Model=QWEN2-VL-7B-INSTRUCT, Alpha (α)=0.03, Candidates=up to two cached candidates, Basis Function Φ(X)=[1, T(X), T(X)^2]2026.03 | 97.27 | 1.42 | 95.21 | |
| TopKBase Model=QWEN2-VL-7B-INSTRUCT, Alpha (α)=0.03, Candidates=up to two cached candidates, Basis Function Φ(X)=[1, T(X), T(X)^2]2026.03 | 96.37 | 1 | 93.23 | |
| Learnt CPBase Model=QWEN2-VL-7B-INSTRUCT, Alpha (α)=0.03, Candidates=up to two cached candidates, Basis Function Φ(X)=[1, T(X), T(X)^2]2026.03 | 96.16 | 1.22 | 94.48 | |
| CFCBase Model=QWEN2-VL-7B-INSTRUCT, Alpha (α)=0.03, Candidates=up to two cached candidates, Basis Function Φ(X)=[1, T(X), T(X)^2]2026.03 | 95.81 | 0.99 | 93.23 | |
| ICPBase Model=QWEN2-VL-7B-INSTRUCT, Alpha (α)=0.03, Candidates=up to two cached candidates, Basis Function Φ(X)=[1, T(X), T(X)^2]2026.03 | 95.58 | 1.84 | 85.21 |