-
Notifications
You must be signed in to change notification settings - Fork 21
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
Can it work without the mask images? #13
Comments
Can you please provide more information about the new dataset (if possible)? Is it similar to datasets like MVTec-AD and VisA, or more like ImageNet and CIFAR-10? I might need more information to assess this. |
This is a medical dataset (images of brain lesions) that has been used for anomaly detection tasks before, but without the information on masks for the abnormal regions. |
The training process is only normal images, testing is normal and abnormal images |
Since our method is designed for zero-shot applications, you can attempt testing it on your dataset by simply changing the text prompt. However, due to the significant disparity between medical tasks and industrial anomaly detection, direct testing may not yield satisfactory results. Therefore, you can also consider retraining our method by generating synthetic anomalies (e.g. DRAEM) to better align with your specific application. The above is just my suggestion. Since I'm not specialized in medical tasks, my ideas may not be entirely suitable. Please take them as a reference. :) |
Thank you so much for your patient explanation and response. I have one more small request: could you please share the manual download link for the CLIP model ( ViT-L-14-336)? |
Sure! You can find all the links you need here. |
Thanks so much. |
I want to use this perfect work in a new dataset, but the dataset doesn't have mask labels. My objective is to perform image-level anomaly detection (normal or abnormal classification). Is that possible to achieve this using this code?
The text was updated successfully, but these errors were encountered: