Fine-grained attribute prediction on RealQA benchmark leaderboard · SOTA2 Research

Fine-grained attribute prediction on RealQA

0.711Eye-Catch SRCCFine-tuned Qwen2-VL-7B (Ours)

Evaluation Results

Method	Links
Fine-tuned Qwen2-VL-7B (Ours)2025.03		0.711	0.755	0.825	0.854	0.493	0.491	0.552	0.641	0.38	0.38	0.405	0.405	0.64	0.702	0.535	0.553	0.752	0.774	0.588	0.617	1
GPT-4o2025.03		0.702	0.722	0.795	0.812	0.232	0.231	0.337	0.372	0.416	0.412	0.238	0.238	0.625	0.676	0.52	0.513	0.633	0.655	0.5	0.515	2
GPT-4o-mini2025.03		0.678	0.707	0.753	0.791	0.164	0.167	0.327	0.341	0.354	0.356	0.266	0.266	0.525	0.549	0.293	0.313	0.736	0.763	0.455	0.473	4
GPT-4V2025.03		0.656	0.708	0.713	0.755	0.212	0.213	0.362	0.397	0.363	0.368	0.319	0.319	0.5	0.576	0.282	0.311	0.585	0.6	0.444	0.472	5
Qwen-VL-Max2025.03		0.648	0.694	0.701	0.749	0.231	0.232	0.461	0.488	0.381	0.376	0.273	0.273	0.581	0.622	0.348	0.361	0.683	0.686	0.479	0.498	3
Qwen2-VL-72B2025.03		0.633	0.703	0.615	0.677	0.263	0.264	0.455	0.476	0.123	0.122	0.248	0.248	0.587	0.645	0.336	0.351	0.652	0.665	0.435	0.461	6
Qwen2-VL-7B2025.03		0.627	0.692	0.616	0.663	—	—	—	—	0.269	0.265	—	—	0.444	0.518	0.171	0.191	0.507	0.559	—	—	7