tiangexiang / SQUID

[CVPR 2023] Deep Feature In-painting for Unsupervised Anomaly Detection in X-ray Images
MIT License
90 stars 6 forks source link

how to segment the lesion in X-ray images rather than giving the image-level prediction? #5

Open cannonli7 opened 10 months ago

cannonli7 commented 10 months ago

Thanks for your work. could you tell me how to segment the lesion in X-ray images rather than giving the image-level prediction?

tiangexiang commented 10 months ago

Hi, thanks for your interests in our work. At the moment, SQUID doesn't output the segmentation mask directly. However, you can rely on the grad-CAM map from the trained discriminator to indicate which parts of the image are poorly reconstructed, and therefore more likely to contain lesions.

cannonli7 commented 10 months ago

Hi, thanks for your interests in our work. At the moment, SQUID doesn't output the segmentation mask directly. However, you can rely on the grad-CAM map from the trained discriminator to indicate which parts of the image are poorly reconstructed, and therefore more likely to contain lesions.

Thanks for your reply. I'll try to work on that.