Aesthetic scores shown in the following figures are in the range [-1,1]. We use the bold font to highlight positive feedbacks and underlines to highlight negative ones.

Expert 2: The aesthetics of these two pictures are mainly affected by the vividness of color and composition. The photo was taken in early autumn, and I think the golden hue can better foreground the characteristics of autumn. Therefore, I like the photo shot before the guidance of Tumera+. Admittedly, the model reasonably capture the important aspect of this picture. For example, color vividness has a large weight for both pictures and it thinks the one after guidance if more vivid. Different from me, the model assigns a larger weight to color harmony and gives a relative low score. It makes some sense, but it ignores the context of this picture.

Volunteer: The app told me that my composition is symmetry and it makes my picture look good. I mean, when I was shooting the first photo. It also said that I should improve the color. I adjusted, but I got a picture with a lower score. This makes me confused. Well, to be honest, I think these two photos are both good-looking.

Analysis: This sample reveals a weakness of our model -- it does not takes into consideration the context in which photos are taken. For the photo shot after guidance, it assigns a low score with a large weight to the light, possibly because it is too bright. In comparison, it gives a high score with a large weight to color harmony for the photo before guidance. However, if it knows that audience typically prefer golden hue in autumn, the model may agree that the color in the photo after guidance is better.

Expert 1: I am of the opinion that there are two problems with the photo shot before guidance, the content of the depth of field. The object is relatively small in the picture and is not very prominent without a good depth of field. The model exactly reflect these aspects. As we can see, the content and depth of field has the lowest scores.

Volunteer: I like Tumera+. It tells me that I could improve my depth of weight and make my content more salient. The photo taken after guidance is better.

Image pair 6

Expert 3: The same as the model, I think the picture shot after guidance is better. The object is more prominent and the composition makes more sense. The color of the foreground and the background is a little bit close, which is also detected by the model, as the two attribute scores regrading color is low for both pictures. However, I do not quite agree that repetition is the factor that decides the aesthetic value of these two pictures.

Volunteer: Actually, the app gives reasonable guidance, and the prompt is timely. But, I think the guidance could be more detailed. For example, how could I improve the repetition of the picture? Though, in the app's defense, I think it needs more context information to give very insightful suggestions.

Image pair 8

Expert 2: Although the second picture conforms the rule of thirds, there are too many objects in the photo and the color is mixed. Consequently, the main object is not salient. One can take advantage of the depth of field to improve the quality of this picture. Since the model assigns very low scores to color harmony and depth of field, I think the analysis it provides is reasonable.

However, there are also some problems with the aesthetic evaluation. First, the attentional map of depth of field is not satisfactory. It highlights the main object and part of the background. In my opinion, highlighting the whole background is better. Second, it assigns a high value (0.21) to symmetry. However, I do not think this picture is symmetric.

Volunteer: The app reminds me that my first picture is too vivid. I think it is a good suggestion, but the app did not give detailed instructions on how to improve it. I had to guess by myself and reduced the content in the picture. I felt good when the app gives me a higher score after I adjusted my picture.