Spaces:
Running
Running
File size: 2,775 Bytes
b650828 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 |
\begin{table}[h]
\centering
\caption{The detailed evaluation result of all multimodal judges on \textbf{quality} perspective. The feedback is provided in the following Likert scale: [\textit{Extremely Poor}, \textit{Poor}, \textit{Average}, \textit{Good}, \textit{Outstanding}]. Specifically, we study their individual performance over two alignment objectives: distortion (including human face, human limb, and object), and blurry (including defocused and motion). The best performance across all models is bolded.}
\resizebox{1.0\linewidth}{!}{%
\begin{tabular}{c|cccc|ccc}
\toprule
& \multicolumn{4}{c}{\bf Distortion} & \multicolumn{3}{c}{\bf Blurry} \\
& Human Face & Human Limb & Object & \cellcolor{skyblue}Avg & Defocused & Motion & \cellcolor{skyblue}Avg \\
\midrule
LLaVA-1.5-7b$^\heartsuit$ & 0.00 & 0.00 & 0.00 & \cellcolor{skyblue} 0.00 & 1.80 & 10.6 & \cellcolor{skyblue} 6.50 \\
LLaVA-1.5-13b$^\heartsuit$ & 0.00 & 0.00 & 0.00 & \cellcolor{skyblue} 0.00 & 18.7 & 29.7 & \cellcolor{skyblue} 24.9 \\
LLaVA-NeXT-mistral-7b$^\heartsuit$ & 10.8 & 14.2 & 1.30 & \cellcolor{skyblue} 9.10 & 56.7 & 73.0 & \cellcolor{skyblue} 61.3 \\
LLaVA-NeXT-vicuna-13b$^\heartsuit$ & 19.6 & 14.3 & 13.9 & \cellcolor{skyblue} 16.8 & 25.8 & 27.3 & \cellcolor{skyblue} 26.6 \\
Instructblip-7b$^\heartsuit$ & 9.80 & 3.00 & 18.7 & \cellcolor{skyblue} 10.9 & 9.80 & 9.90 & \cellcolor{skyblue} 9.50 \\
Prometheus-Vision-7b$^\heartsuit$ & 19.8 & 15.6 & 12.2 & \cellcolor{skyblue} 16.0 & 26.0 & 29.2 & \cellcolor{skyblue} 27.2 \\
Prometheus-Vision-13b$^\heartsuit$ & 7.40 & 5.10 & 7.30 & \cellcolor{skyblue} 6.80 & 9.40 & 11.7 & \cellcolor{skyblue} 11.1 \\
Qwen-VL-Chat$^\spadesuit$ & 25.2 & 21.6 & 6.70 & \cellcolor{skyblue} 17.4 & 18.8 & 20.1 & \cellcolor{skyblue} 19.3 \\
Internvl-chat-v1-5$^\spadesuit$ & 22.1 & 24.2 & 1.20 &\cellcolor{skyblue} 16.0 & \bf 94.2 & 96.1 & \cellcolor{skyblue} \bf 95.3 \\
Idefics2-8b$^\spadesuit$ & 40.9 & 29.6 & 10.1 & \cellcolor{skyblue} 27.0 & 90.2 & 67.5 & \cellcolor{skyblue} 79.2 \\
\midrule
GPT-4-vision$^\clubsuit$ & 86.9 & 54.4 & 78.7 & \cellcolor{skyblue} 71.5 & 90.6 & \bf 93.5 & \cellcolor{skyblue} 93.6 \\
GPT-4o$^\clubsuit$ & \bf 98.2 & \bf 71.1 & \bf 89.9 & \cellcolor{skyblue} \bf 83.6 & 91.8 & 96.1 & \cellcolor{skyblue} 91.6 \\
Gemini Ultra$^\clubsuit$ & 71.3 & 30.5 & 59.2 & \cellcolor{skyblue} 48.8 & 80.6 & 90.9 & \cellcolor{skyblue} 79.5 \\
Claude 3 Opus$^\clubsuit$ & 21.3 & 17.2 & 9.50 & \cellcolor{skyblue} 14.0 & 85.9 & 93.1 & \cellcolor{skyblue} 83.7 \\
\bottomrule
\end{tabular}%
}
\label{exp:artifact_result_narrative_5}
\end{table}
|