search

yulunzhang / RDN

Torch code for our CVPR 2018 paper "Residual Dense Network for Image Super-Resolution" (Spotlight)
543 stars 108 forks source link

model does not converge #1

Closed echolijinghui closed 5 years ago

echolijinghui commented 6 years ago

hi，sorry to bother you,but I have a problem that model does not converge,what should I do?can you give me some suggestions?thank you so much.the loss function i used is l1 loss

default

yulunzhang commented 6 years ago

Hi,

Did you run the provided torch code? What training script did you use?

If we train the network from scratch (e.g., for BI, X2), the loss should converge fast at the beginning. bix2_loss

If we fine-tune the network based on previous trained models, the loss might fluctuate at the beginning. After the learning rate decreases, the loss would also trend to decrease. bix2_finetune_loss

However, no matter we train from scratch or fine-tune, the L1 loss value should not be so large. So, could you give more training details about your posted results? Thanks.