Spaces:

gokaygokay

/

Florence-2

like 661

Running on Zero

I think the code should be changed like this

    if text_input is None:
        prompt = task_prompt
    else:
-        prompt = task_prompt + text_input
+        prompt = task_prompt + " " + text_input

for example,

test image: http://farm3.staticflickr.com/2386/2532343535_41a2d3a9a0_z.jpg (from coco)
task prompt: Region to Description
text input: man on the back (without space)

output:

{'<REGION_TO_DESCRIPTION>': 'A woman with a large backpack in an airport terminal.'}

text input: man on the back (with prepending a space)
output:

{'<REGION_TO_DESCRIPTION>': "person on the back of a large green backpack with straps and buckles. \n\nThe backpack appears to be made of a durable material and has multiple pockets and compartments for storage. The straps are adjustable and the buckles are silver. The backpack is resting on a blue and white checkered floor.\n\nThere is a person's leg visible on the right side of the image, but they are not clearly visible. The background is blurred, so it is difficult to make out any other details."}

I think the example code from microsoft/Florence-2-large is wrong.