search

Tombs98 / CAPTNet

Prompt-based Ingredient-Oriented All-in-One Image Restoration
https://arxiv.org/abs/2309.03063
Other
17 stars 0 forks source link

About the training #4

Open mawei-north opened 1 month ago

mawei-north commented 1 month ago

Hello, I'm intrigued by your work! Are you putting all the task dataset images in the same folder during training? How was data equalization implemented in your paper?

Tombs98 commented 1 month ago

Hello, I'm intrigued by your work! Are you putting all the task dataset images in the same folder during training? How was data equalization implemented in your paper?

Yes, in the same folder. In order to maintain the approximate amount of data for each degradation type, we use data enhancement methods such as flipping and so on.