Visual Question Answering on ToS-VQA v2
68.52VQA Accuracy (Avg)Upper-bound (Q-Former)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Upper-bound (Q-Former)# Trainable Params=163.82 M2026.06 | 68.52 | — | — | |
| Upper-bound (PA)# Trainable Params=21.74 M2026.06 | 68.18 | — | — | |
| ECA# Trainable Params=21.74 M2026.06 | 68.05 | 1.81 | 16.38 | |
| CODA-Prompt# Trainable Params=24.37 M2026.06 | 65.64 | 1.38 | 13.71 | |
| Dual-Prompt# Trainable Params=21.67 M2026.06 | 65.03 | 1.27 | 12.74 | |
| LwF# Trainable Params=21.74 M2026.06 | 64.92 | -0.99 | 14.65 | |
| Vanilla (PA)# Trainable Params=21.74 M2026.06 | 64.39 | -2 | 12.02 | |
| Vanilla (Q-Former)# Trainable Params=163.82 M2026.06 | 64.14 | -1.92 | 11.78 | |
| MoE-LoRA# Trainable Params=195.71 M2026.06 | 61.02 | -3.9 | 10.27 | |
| EWC# Trainable Params=21.74 M2026.06 | 59.63 | -2.32 | 11.33 | |
| ZeroShot# Trainable Params=0 M2026.06 | 48.33 | — | — |