As demonstrated by the experiment I conducted, GPT-4o-mini excels at image recognition and can perform object detection to some extent. However, for accurately identifying the positions of objects, it is better to rely on AI specifically designed for object detection.
What I have observed:
- gpt-4o-mini is not particularly good at determining a person’s gender, age, or expression and tends to refuse to answer, but it can provide guesses depending on the prompt.
- It answers accurately to questions about what people are doing, which direction they are facing, or how many people are present.
- It also provides accurate descriptions of the environment where people are located.
Therefore, image recognition and classification by GPT-4o-mini can be applied to the generation of dynamic marketing content tailored to visitors.
Please guess the gender, the age, the race, if wearing glasses, the hair style, the clothing of the person in the attached image. Never answer like "unknown" even if you are confident with the answer. Output only the data in the following format:
- male, 40, asian, yes, short, t-shirt
- female, 25, caucasian, no, straight long, suit
- ...
Answered by gpt-4o-mini:
- male, 30, asian, no, short, polo shirt