DFR
DFR copied to clipboard
YoungGod
Project: Unsupervised Anomaly Segmentation via Deep Feature Reconstruction
hi, during evaluation, how can I get the amomaly score map and the anomaly segmentation image?
Thanks for your code. Can you tell me what autoencoder.pth is? how to
In training, is the L2 distance used in the loss function and the relevant Euclidean distance not used? Is the experimental result of L2 distance better? Thanks!
The results I got are quite different from your paper in some categories, Can you show your torch version, python version and so on?
Hi, Thanks for your code!! extractor was not included in the optimizer parameter. So, I think the extractor was only used for inference, but wonder why used train(). In this...
``` class TestDataset(Dataset): ...... def _get_image_files(self, path, ext={'.jpg', '.png'}): images = [] for root, dirs, files in os.walk(path): if "check" in root: continue print('loading image files ' + root) #...
I use the mask and anomaly_map of metrics_evaluation fuction write a out_image,but it is different with paper? Can you help me solve the problem?
What are the specific parameter Settings
