search

ytongbai / LVM

Apache License 2.0
1.77k stars 55 forks source link

Question of Fig. 13 in the Paper #3

Open Unrealluver opened 12 months ago

Unrealluver commented 12 months ago

Greetings,

Thanks for your awesome work! This work shed light on the AGI sparks with pure-vision models! Reading your paper, I found Fig. 13 about IQ testing amazing. Does the training set include similar images, i.e., IQ testing images? That is an important question about the reliability of AGI for this paper.

JingfengYao commented 12 months ago

Thank the authors for the great work!

The same question. Are there some IQ-testing-like images in the training dataset? If not, the performance is really amazing!

mattdeitke commented 12 months ago

I think they're mocking the sparks of AGI paper in Figure 13 (given the 🤨 emoji) 🙂.

I think many inpainting methods can already do these tasks. For instance, here's the example of using generative fill in photoshop, with 3 different generations:


Any inpainting method has to reason about its surroundings and fill in the blank based on context :)

Unrealluver commented 12 months ago

I think they're mocking the sparks of AGI paper in Figure 13 (given the 🤨 emoji) 🙂.

I think many inpainting methods can already do these tasks. For instance, here's the example of using generative fill in photoshop, with 3 different generations:

Any inpainting method has to reason about its surroundings and fill in the blank based on context :)

Great！Thanks for trying that! Could you please try more instances with Photoshop, e.g., the end image of the first row and third row?

ytongbai commented 11 months ago

Hi yes, this was a friendly joke and we'll check if we can do more interesting cases later. Thank you all for your kind support and interests, we really appreciate it.

Best, Yutong