Image maximum resolution
Hello,
I noticed that text recognition in images works extremely poorly for very high-resolution images (8k), scaling down to 1k gives me perfect results. I'm running the model with vLLM. Should this be automatically resized? What resolution does the model natively support?
thanks for this wonderful model!
Unless specified otherwise, most models (this one included) run on 1K resolution. They have no idea how to process a 8K image. Normally, good frontends will resize that for you, but if you're using the model more "directly", yep, that's your job to do so.
oh perfect thank you! yeah openwebui seems to ignore it, but i´ve already implemented the resizeing in my interface.
Also funny things will happen when the image is rotated, but file explorer shows it correctly due to the metadata :D
Thanks a lot!