search

Xiaofeng-life / SFSNiD

30 stars 1 forks source link

Hello, how to get the pre-training weights ("last_SFSNiD.pth") in step2？ #4

Closed chener311 closed 2 months ago

Xiaofeng-life commented 4 months ago

The pre-trained model is not needed in Step 2. You can train the model from scratch.

chener311 commented 4 months ago

Thanks for the answer. In addition, the picture became blurry after training, do you have such a situation?

Xiaofeng-life commented 4 months ago

Thanks for the answer. In addition, the picture became blurry after training, do you have such a situation?

The results (pictures) obtained by the supervised training process should be with blurry details. Which dataset set you are using?

chener311 commented 4 months ago

I used the NightHaze dataset, and both the semi-supervised training and the final result were blurrier than the original hazy image.

Xiaofeng-life commented 4 months ago

I used the NightHaze dataset, and both the semi-supervised training and the final result were blurrier than the original hazy image.

NightHaze dataset do not need semi-supervised traning. The supervised process is designed for it. The semi-supervised process is used for real-world images.