Reconstructing the problem of image preservation

cnulab / RealNet

Offical implementation of "RealNet: A Feature Selection Network with Realistic Synthetic Anomaly for Anomaly Detection (CVPR 2024)"

MIT License

237 stars 14 forks source link

Reconstructing the problem of image preservation #72

Open yangjq321 opened 1 week ago

yangjq321 commented 1 week ago

Hello, this work of yours is fantastic! A great help for me. I would like to ask, how to save the reconstructed image in the test phase evaluation_realnet.py?

cnulab commented 6 days ago

RealNet reconstructs in the feature space of the image, so the reconstructed image cannot be obtained directly. If you need to visualize the reconstructed image, it may require training an additional decoder.