Each screen shows a short description and two generated images, A and B, made from the same starting noise by two different models. Which model made which image is hidden and changes from pair to pair.
Relation holds: the first object is placed as described relative to the second (left of, right of, above, below). Judge only the placement.
Looks better: overall image quality: sharper, more realistic, fewer distortions. Ignore the description.
The model is still in training, so objects can look rough: a duck may be just a yellow blob. Judge roughly.
You may skip a question you cannot judge and still press Next.
You will see about 27 pairs, which takes around 7 minutes. No sign-in is needed. Your answers are sent automatically when you finish. Progress is kept in this browser, so you can pause and come back on the same device.
Connecting…
Descriptiona stool below a piano
A
B
1. In which image does the relation hold? (is the first object placed as described?)
2. Which image looks better? (sharper, more realistic, fewer distortions; ignore the description)
Enter = next. You may skip a question you cannot judge.
This set is done. Thank you.
Backup: if the line above says sending failed, copy this code and send it to the person who shared the link.
If you have another 10 minutes, you can rate a second set of different pairs; the code then covers both sets.