search

marco-rudolph / cs-flow

This is the official repository to the WACV 2022 paper "Fully Convolutional Cross-Scale-Flows for Image-based Defect Detection" by Marco Rudolph, Tom Wehrbein, Bodo Rosenhahn and Bastian Wandt.
113 stars 29 forks source link

n_scales parameter #19

Closed CharlesAntoineParent closed 2 years ago

CharlesAntoineParent commented 2 years ago

Hello ! I was wondering why there was an option to configure the number of scale produce by the feature extractor if nf blocks are always supposed to get 3 scales ? Thank you

marco-rudolph commented 2 years ago

Hi, unfortunately, there is currently no option to configure the number of scales. n_scales is used for many parts of the code for readability, but not for the NF architecture itself.

Best regards, Marco Rudolph

CharlesAntoineParent commented 2 years ago

Thank you for the quick answers. I was wondering if you tried getting multi scale representation by using a backbone that produces different scales only with one propagation (like Darknet + Path aggregation network in yolo) instead of getting representation by doing multiple propagation in the backbone ?

marco-rudolph commented 2 years ago

I tried this, but got a slightly worse performance for MVTec. It might be beneficial for other datasets.