Visual Question Answering on InstaOrder Occ. (val)
85.2AccuracyQwen3-VL-4b-Instruct
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Qwen3-VL-4b-Instruct2026.03 | 85.2 | — | 49.6 | 2.7 | 5.1 | |
| SNs (maj. voting)Backbone=LLaVA-v1.5-7b2026.03 | 78.2 | 62 | 30.4 | 36.5 | 33.2 | |
| SNs (mean)Backbone=Qwen3-VL-4b-Instruct2026.03 | 78.1 | 62 | 31.3 | 40 | 35.1 | |
| SNs (maj. voting)Backbone=Qwen3-VL-4b-Instruct2026.03 | 77.2 | 62 | 29.2 | 37.8 | 33 | |
| LLaVA-v1.5-7b2026.03 | 64.9 | — | 18.5 | 40.2 | 25.3 | |
| SNs (mean)Backbone=LLaVA-v1.5-7b2026.03 | 64.5 | 62 | 23.5 | 61.8 | 34 |