Add Evaluation Task 1 Instruction Text

[huangziwei]

Picking the fake image from three images. Images will disappear after 5 seconds.

[induilanchezian]

Task 1 instructions:

• For each question in this task, you are shown a set of three fundus images or OCT scans labelled as a, b, and c.
• You have to determine which image is fake i.e. generated by our deep learning model and select the label corresponding to the generated image.
• You will have a time limit of x s for each question. Images will be hidden after the completion of the time limit.
• You can zoom in to observe the image regions more closely to make your decision.

[huangziwei]

There's no magnifier for task 1 currently, since it's a time-limited task, wouldn't it be too distracting if we allowed user to zoom in?

[philippberens]

The title should be "Realism of AI-generated fundus images"

Instructions:

• You will be shown a set of three fundus images scans labelled as a, b, and c.
• The images will be shown for 20 s. The images will be hidden after the completion of the time limit.
• In each panel, two of these images have been taken from real patients, one has been generated by an AI.
• Select the label corresponding to the image generated by the AI.

Information about the study: Your answers will be saved on a secure server in the EU and will be used to assess the quality of image generation of an AI algorithm. If you have additional questions, do not hestiate to ask. No personal information or information about the quality of your answers will be disclosed at any time to anyone, in particularly it will not be shared with your employer. Please check "consent" below if you agree to these terms.

[philippberens]

@induilanchezian If we also run it on OCTs, we should add that as a separate task, and adjust the wording respectively.

[induilanchezian]

Instructions edited for OCT page: Title: "Realism of AI-generated Optical Coherence Tomography (OCT) images"

Instructions:

• You will be shown a set of three OCT images labelled as a, b, and c.
• The images will be shown for 20 s. The images will be hidden after the completion of the time limit.
• In each panel, two of these images have been taken from real patients, one has been generated by an AI.
• Select the label corresponding to the image generated by the AI.

Information about the study: Your answers will be saved on a secure server in the EU and will be used to assess the quality of image generation of an AI algorithm. If you have additional questions, do not hestiate to ask. No personal information or information about the quality of your answers will be disclosed at any time to anyone, in particularly it will not be shared with your employer. Please check "consent" below if you agree to these terms.