search

Yuehan717 / PDD

[ECCV 2024] The Official Code for "Pairwise Distance Distillation for Unsupervised Real-World Image Super-Resolution"
MIT License
17 stars 1 forks source link

data degradate too slow #2

Open zhuzhu18 opened 2 weeks ago

zhuzhu18 commented 2 weeks ago

训练速度太慢了，建议像RealESRGAN一样将真正执行的降质操作放在SRGANRefModel的feed_data中，同时用队列来打乱每个batch的样本，维持降质多样性，这样应该可以对训练加速不少