Results without guidance #46
Description
Congrats on the great paper and impressive results!
I noticed that the current results primarily utilize guidance. Since it's standard practice in the field to provide baseline results without guidance, using CFG mainly for complementary analysis, the lack of "no-guidance" metrics makes it difficult to compare this work against other methods.
A few quick questions:
- Does the performance of JiT significantly drop without guidance?
- Is there a specific reason why guidance appears particularly critical for this pixel-model architecture?
Providing results without guidance would greatly help in situating this work within the existing literature.
Thanks!