Closed alexander-fischer closed 9 months ago
Hi, the iteration over patch_tokens
is carried out to individually compare the stored normal features and test images of different blocks (which is 4 in the code).
Besides, during data preparation, we employ the variable k_shot
, and for example, the names of the randomly selected reference images are recorded in ./results/mvtec/few_shot/4shot/seed42/k_shot.txt
.
You're right!
From debugging the code, it seems that here the patch tokens of just one reference image are compared to the query image since we iterate over
patch_tokens
that has the length offew_shot_features
. Shouldn't it be possible to use multiple reference images with thek_shot
variable?https://github.com/ByChelsea/VAND-APRIL-GAN/blob/46fcbe567752dbcc84a26a05fb7b3794a72aa008/test.py#L189-L201