RealBasic video super resolution 足球场地画面

We have trained the RealBasicVSR 2X model by ourselves using the official method. After the over-splitting, the scene has a lot of noise and the picture effect is not improved. Can we modify some parameters of the script for adjustment 我们使用官方的方法,自己训练了RealBasicVSR 2X的模型,超分后场地画面画面噪点较多,画面效果没有提升,是否可以修改脚本的一些参数进行调整
We recommend using English or English & Chinese for issues so that we could have broader discussion.
Hi,
Can you please upload the training logs for our reference?
https://github.com/open-mmlab/mmediting/issues/908#issuecomment-1144479723 this is training logs ,Please provide modification suggestions
Hello I haven't tried x2 models before. How many frames did you use during inference?
- By modifying the scale values of RealBasicVSR_WOGAN_C64B20_2X30X8_LR1E-4_300K_REDS and RealBasicVSR_C64B20_1X30X8_LR5E-5_150K_REDS, Retraining using REDS 'Train_Sharp dataset has approximately 1110 images
- May I ask whether the above method is feasible or whether the model effect can be optimized by adjusting parameters
- for example modify RealBasicVSR_WOGAN_C64B20_2X30X8_LR1E-4_300K_REDS and diff

@Z-Fran Please check this issue.