Qwen

/

Qwen2.5-VL-7B-Instruct

like 1.34k

Follow

Qwen 57.2k

Hi,
I am trying to use Qwen 2.5-VL for classification of microscopy images. My images are 16-bit grayscale and I am using crops of 110x110 pixels. For each microscope image I actually have 9 dimensions (3 z-planes * 3 fluorescent channels). I know that the input expects 8-bit RGB images. What would be the best way to input my images into Qwen? Should I group my images into 3 "fake RGB" images (i.e. the channels aren't the real colors)? Is there a suggested prompt I can use to insert several images together which will inform the model that each image belongs to different z-plane and channel?

Thank you in advance!