not funny to general images
Opened this issue · 1 comments
I can rarely find funny responses, can it generalized well to a broader range of images?Creativity is challenging. The images used in the Oogiri game are general images. In fact, as mentioned in our paper, innovation itself is a tough endeavor. Even top human players find it challenging to come up with responses for general images in the Oogiri game (such as video: link 19:42~28:57). Therefore, it's hard to expect models to generate satisfying responses for any given image. Despite there may exist images where finding a funny response is nearly impossible, the creative humorous responses we've seen so far demonstrate the effectiveness of our model.
Humor is so subjective. For example, your generated response "最近一直加班，每天晚上都在加班" is a dark humor that thinks outside the box for me while maybe you don't think it funny. I think it fits well with the image of deserted streets and the description of someone looking weary from frequent late-night work. We suggest trying out more generations to find responses that suit your preferences.
I've also made some attempts for your image, and here are the results. At least these model outputs are what I think interesting.
