Generated inlier/outlier embedding seems not being used when synthesizing images

[wangpuyi]

Thanks for your great work!

However, when trying to synthesiszng images in the pixel space following your guidance, I find it seems that loaded_embedding is never used in the script and condition is prompted as line.

It shows that random noise is added to change condition instead of using generated inlier/outlier embedding in the previous stage, which confuses me a lot and hope to get your explanation.

Thanks for your consideration.

[LarsDoorenbos]

I am also curious about this.

[LarsDoorenbos]

Found it: it's hidden here

dream-ood/ldm/modules/encoders/modules.py

Line 310 in ca647e3

self.outlier_embedding = torch.from_numpy(