search

ByChelsea / VAND-APRIL-GAN

[CVPR 2023 Workshop] VAND Challenge: 1st Place on Zero-shot AD and 4th Place on Few-shot AD
162 stars 20 forks source link

Query about zero-shot anomaly segmentation #6

Open BehzadBozorgtabar opened 11 months ago

BehzadBozorgtabar commented 11 months ago

As far as I understood, unlike WinClip for anomaly segmentation, the original ground truth of test images is used for model adaptation, while the final performance is reported on the same test images. Can you please confirm?

ByChelsea commented 11 months ago

Hi, our method is trained on Dataset A (e.g., MVTec AD) and tested on Dataset B (e.g., VisA). It is worth noting that although we use the test set of Dataset A during training, the training and testing are not conducted on the same dataset.