search

BinWang28 / AnomalyHop

IEEE VCIP 2021: AnomalyHop: An SSL-based Image Anomaly Localization Method
MIT License
14 stars 2 forks source link

Working with non-square images #1

Closed alevangel closed 2 years ago

alevangel commented 2 years ago

Hi, I see that in the loading function of the dataset there is a crop transform. Why this is happening? It would be possibile to work with a non square images dataset?

BinWang28 commented 2 years ago

Yes, it can be adapted to different input sizes.

By doing this, we follow the same processing steps and inputs to have a fair comparison with previous work on the same dataset. https://github.com/xiahaifeng1995/PaDiM-Anomaly-Detection-Localization-master/blob/main/datasets/mvtec.py

Let me know if you have further questions.