search

cqylunlun / GLASS

[ECCV 2024] Official Implementation and Dataset Release for <A Unified Anomaly Synthesis Strategy with Gradient Ascent for Industrial Anomaly Detection and Localization>
MIT License
156 stars 19 forks source link

About Custom dataset #8

Closed wragon closed 3 months ago

wragon commented 3 months ago

If my custom dataset doesn’t have ground truth data and therefore no mask_path, how do I train?

cqylunlun commented 3 months ago

In the paper, we follows the general paradigm of anomaly detection methods. Here is a feasible solution for practical application:

  1. Modify mvtec.py to set maskpaths_per_class to None.
  2. Comment out the code for metric calculation and model selection in glass.py.
  3. Select the model from the current or final epoch.