From Hindsight to Foresight: Self-Encouraged Hindsight Distillation for Knowledge-based Visual Question Answering

Benchmark Model Rank Results
visual-question-answering-on-ok-vqaHinD-CoT-Know (InternVL3-8B)–Accuracy: 68.6
visual-question-answering-on-ok-vqaHinD-CoT-Know (Qwen2.5-VL-3B)–Accuracy: 65.9
visual-question-answering-on-ok-vqaHinD-CoT-Know (Qwen2.5-VL-7B)–Accuracy: 67.5
visual-question-answering-on-ok-vqaHinD-Know (Qwen2.5-VL-3B)–Accuracy: 64.7
visual-question-answering-on-ok-vqaHinD-Know (Qwen2.5-VL-7B)–Accuracy: 68.3