search

jingyuanli001 / RFR-Inpainting

The source code for CVPR 2020 accepted paper "Recurrent Feature Reasoning for Image Inpainting"
MIT License
355 stars 76 forks source link

What does fintune mean? #25

Closed laolongboy closed 2 years ago

laolongboy commented 3 years ago

Thanks for your contributions. 'For CelebA Dataset, train the model for 350,000 iterations and finetune for 150,000 iterations. (500,000 in total)' I want to know what's the meaning of finetune in the same dataset. According to the code, when finetuning, parameters of BatchNorm2d will be fixed and the lr will change from 2e-4 to 5e-5. Do I ignore some other operations ?

jingyuanli001 commented 3 years ago

Your understanding regrading fine-tuning is completely correct. In fact, the main purpose of fine-tuning here is to stabilize the batch normalization layers which is not stable in inpainting tasks, due to the unpredictable mask shape and region. Without this procedure, the image color might be distorted.

jingyuanli001 commented 3 years ago

You might refer to #33 to see what do the images look like before freezing the batch normalization.