6. User-Study of entity-Level Text-Guided Image Manipulation

We will provide some generated images in each question and we hope you can rank these images from the following perspectives:
(1) Consistency: Do the generated images match the text description?
(2) Fidelity: Do the generated images resemble real images?
(3) Preservation: Have the generated images maintained the background?
Ranking 1 is the best and ranking 5 is the worst.

Rank the following images from the perspective of Consistency: Do the generated images match the text description?
Description：a grey bird with a white belly and breast and small bits of white on its wings.[Please sort the items]

Option 3
Option 5
Option 4
Option 1
Option 2

Rank the following images from the perspective of Fidelity: Do the generated images resemble real images?[Please sort the items]

Option 3
Option 2
Option 1
Option 5
Option 4

Rank the following images from the perspective of Preservation: Have the generated images maintained the background?
Option 6 is the original image. Please DO NOT select it when ranking.[Choose at most 5 items and sort them]

Option 4
Option 1
Option 5
Option 3
Option 2
Original image. DO NOT select.

Rank the following images from the perspective of Consistency: Do the generated images match the text description?
Description：this flower has petals that are blue and very thin[Please sort the items]

Option 3
Option 5
Option 2
Option 1
Option 4

Rank the following images from the perspective of Fidelity: Do the generated images resemble real images?[Please sort the items]

Option 4
Option 3
Option 5
Option 2
Option 1

Rank the following images from the perspective of Preservation: Have the generated images maintained the background?
Option 6 is the original image. Please do not select it when ranking.[Choose at most 5 items and sort them]

Option 4
Option 1
Option 2
Option 3
Option 5
Original image. DO NOT select.

Report