search

cnulab / RealNet

Offical implementation of "RealNet: A Feature Selection Network with Realistic Synthetic Anomaly for Anomaly Detection (CVPR 2024)"
MIT License
143 stars 5 forks source link

Fine-tuning #5

Open khawar-islam opened 3 months ago

khawar-islam commented 3 months ago

Hello @eltociear @cnulab

Thanks for you great work. Can I fine-tune your model on new defect private dataset which is related with defects? If yes, How we can do it? Because there is no file and you can just give me the hint.

Regards, Khawar

cnulab commented 3 months ago

Thank you very much for your interest in our work. I understand that you want to train (fine-tune) a new anomaly synthesis model on your private dataset. If yes, you first need to reimplement the SDASDataset class in datasets/sdas_dataset.py to ensure the correct loading of your dataset. Then, use your custom Dataloader in train_diffusion.py. I recommend that you fine-tune from the ImageNet diffusion checkpoint, as that might achieve better generalization. If you have any further questions, please feel free to continue the conversation.