ChatGPT The image recognition capabilities, provided by OpenAI's gpt-4o, gpt-4o-mini, and gpt-4-turbo models, perform well in many scenarios, but accuracy is not absolute. Here are the key points that affect its performance:
✨ Areas of specialization:
- Generalized identification: ChatGPT is best at answering questions about the "what" of an image, such as recognizing objects, scenes, and underlying relationships. More specificallyVisual Target Detection, ChatGPT is not good at it.
⚠️ Limitations and Impact Factors:
- Image quality is fundamental:
- Clarity, lighting and occlusion directly affect recognition. Blurring, too dark/too bright, and occlusion of key objects all reduce accuracy.
- Image complexity is the challenge:
- A large number of objects and a complex background can make identification more difficult.
- Level of detail (detail parameter) Controllable: (API interface optional)
- LOW: Fast, low resolution (512x512px), consumes 85 tokens, good for scenes that don't need high detail.
- high：更准确，但速度较慢，消耗更多 tokens（每个 512×512 区域 170 tokens (+85 tokens). Ideal for scenes requiring high detail.
- auto: the model is automatically selected.
- Scenario-specific caution is required:
- Spatial orientation: Not good at precise spatial orientation.
- Medical Images: inapplicableIn Medical Image Interpretation.
- Non-Latin alphabet: Recognition may be poor. (e.g. Chinese, Japanese, Korean)
- Small text/rotation/special styles: Need to zoom in, avoid rotation, and pay attention to line style.
- Panorama/Fisheye: Difficult to deal with.
- Count: The results may be only approximate.
- Captcha and image metadata are not supported
- Image size and cost (API)
- Limit upload size:20MBThe
- Image size expectations for different levels of detail:
* Low-res: 512px X 512px
* High-res: Less than 768px on the short side and less than 2000px on the long side. - Costing:
- Low res: 85 tokens for any size image.
- High res: 会根据图片大小进行缩放，每 512px 方块 170 tokens，再加上85 tokens。例如，1024×1024 的图片，费用为 765 tokens；2048×4096 的图片，费用为 1105 tokens。
💡 Summary:
ChatGPT's image recognition is accurate in many cases, but is affected by a number of factors. For best results, provide clear, high-quality images, select the appropriate level of detail, and be aware of the limitations listed above. More specialized tools may be required for high-precision needs or special image types.
May not be reproduced without permission:Chief AI Sharing Circle " How accurate is ChatGPT image recognition?
English 
简体中文 
日本語 
Deutsch 
Português do Brasil