Is there a way to communicate with any gpt-4 model via an API asking it to identify objects in an image and ask it to return the confidence score of the objects identified ? The point is that it sometimes misses objects or returns similar objects so I want it to return the detected objects with the 2 or 3 highest confidence scores instead of the model anticipated top one only
Normally, this would be where we would recommend logprobs.
However, GPT-4-vision-preview currently does not support this. So, unfortunately, this is not possible. Yet.