EmoGenie
A multimodal benchmark for emotion understanding in AI-generated images. It covers two levels: emotions expressed by the image and emotions evoked in the viewer. It comprises a question-answering understanding part and a continuous scoring part, and is accompanied by the evaluation model EmoProbe.
Timeline (last 90 days)
EmoGenie is currently a rare unified benchmark that simultaneously covers both the emotion expressed by generated images and the emotion evoked in viewers.
EmoGenie is accompanied by an evaluation model, EmoProbe.
The research team used 800 emotion-targeted prompts to drive 8 T2I models to generate 6,400 images.
Experimental results show that EmoProbe performs better on emotion-related QA tasks, and its emotion and aesthetics predictions correlate better with human ratings.
EmoGenie provides a quantifiable testing tool for the emotional controllability of text-to-image models, and can be used to examine the reliability of MLLMs as AI judges.
EmoProbe is currently a method proposed alongside the EmoGenie benchmark, and its generalization ability on other image distributions and other MLLMs still requires more independent validation.
All Facts (6)
EmoGenie is currently a rare unified benchmark that simultaneously covers both the emotion expressed by generated images and the emotion evoked in viewers.
50%UnverifiedEmoGenie is accompanied by an evaluation model, EmoProbe.
50%UnverifiedThe research team used 800 emotion-targeted prompts to drive 8 T2I models to generate 6,400 images.
50%UnverifiedExperimental results show that EmoProbe performs better on emotion-related QA tasks, and its emotion and aesthetics predictions correlate better with human ratings.
50%UnverifiedEmoGenie provides a quantifiable testing tool for the emotional controllability of text-to-image models, and can be used to examine the reliability of MLLMs as AI judges.
50%UnverifiedEmoProbe is currently a method proposed alongside the EmoGenie benchmark, and its generalization ability on other image distributions and other MLLMs still requires more independent validation.
50%