-
Notifications
You must be signed in to change notification settings - Fork 668
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Improving the result for custom dataset #285
Comments
My pixel_AUROC , pixel_F1Score values are always getting me 0. I don't understand where is the problem. |
Hi @ZeynepRuveyda, can you share your |
dataset: model: metrics: project: optimization: |
Hi @samet-akcay , here is my config file. |
My custom images looks like metal_nut images which in MVTech dataset. |
Can you show you folder structure for the training and testing data? Sounds like your ground truth pixel masks aren't in place if you're getting 0 scores only for the pixel metrics. what are your image-wise metric results? |
+1 for this. @ZeynepRuveyda, did you also try other models? If so, did you get 0 for those as well? |
Hi @alexriedel1 ,
─────────────────────────────────── .../zeynep/anomalib/datasets/MVTec/BACK/test/bad/000065.png My ground_truth folder stay with train and test folder in the same directory. |
Yes,I tried the patchcore model too but I got the same result. |
Unfortunately I cannot say if the ground truth masks are named according to your anomaly samples. @samet-akcay maybe the dataloaders need a warning (or info) if no ground truth masks are found.. |
I think i missed the file in the directory. I exactly the same directory for ground_truth what you said. |
Can you post an image of the test inference from |
@ZeynepRuveyda, can you share the full training logs printed on the terminal? If there is any peculiarity in the masks, there could be some warnings there. In addition, I'd also look at the visualized outputs, specifically to the anomaly heatmap. If it is correctly produced, this means the model should be able to successfully perform the segmentation. |
|
Yesterday , I got some result for the segmentation and heatmap. But my pixel_AUROC and pixel_F1Score are still 0. Is it problem? |
Are your ground truth masks binary files only containting 0's and 1's ? |
That's what I suspect as well. I can replicate the issue by using a value other than 1.0 for the anomalous regions in the GT masks. Could you provide an example of one of the ground truth mask files of your dataset? |
I think my annotations are just for the classification. I double checked it then you are right. It can be change depend on your mask info. |
Hi,
I am able to running the code but when I am working on the custom dataset, the code did not achive to detect the defect. I tried change the threshold values ,epoch numbers etc.. Do you have any advice for getting good result on my custom dataset?
Thanks,
The text was updated successfully, but these errors were encountered: