I am trying to populate bounding boxes for a given list of fields, the results are always off from the actual value. Is there any way to fix this issue?
Hi @gauraww,
Putting the same response:
Which model you are using? Try to use the latest model with updated one like 2.5 pro. It also depends on prompting. Do some prompt iteration , refine your prompt and see if you are getting better result. If none of this working, it might be model response issue.
Using Flash 2.0. It give better result if I just asked bounding box for everything, but becomes inaccurate as I increase the context in the prompt (i.e. to make the bounding box for specific fields, or finding a key value pair). Will try 2.5 pro and see if I get better result
Yeah, perfect. Try 2.5 and let see. You can also try chain of thought here and see.
The bounding box does not seems to be improved by 2.5 pro but the response time is around 3x compare to 2.0. Hope this latency gets improved in future.
Since it’s a thinking model it’s taking little longer. Latency with 2.5 is already raised with the team. I recommend using images as of now of bounding boxes.
For the images too, the bounding box is inaccurate. Seems like have to use some other tools along with Flash 2.0 for the bounding boxes.