hadesfgh

compute_flow() error：Error in calculating optical flow of only one frame

4

Recently, I'm trying to use restoration_video_demo.py to test the super-resolution effect, and the video we used have 321 frames in total, max_seq_length=10. In restoration_video_inference.py line 128, the data.size(1)=1, and errors...

Have your team ever tried to compress the super-resolution models, such as realbasicvsr. Or does open-mmlab have any relevant tools to compress thees models.

Sorry to bother you. When using GPU to test the effect of super-resolution, we found a problem. Specifically, we used restoration_video_demo.py to restore a 720p, 45s video. We found that...

The super-resolution effect of subtitles or text in video is poor

2

Recently, we use real-basicvsr network to restore many videos with subtitles. For some subtitles or text in the video, the effect of super resolution is poor. I would like to...

如何重新训练scnn

3

你好，最近我根据您论文中的思路和github代码进行了复现。目前想尝试修改SCNN架构，但是缺少对应的数据集，查阅了相关的代码https://github.com/zwx8981/DBCNN，由于能力有限，没看懂如何生成训练SCNN预训练模型的数据集（包括失真图片和label），您能否提供python版本生成预训练数据集的代码，不胜感激。 还有一个问题，对于DBCNN的优化，您是否有一些思路。

Thank you for your great work. Seeing your question [6](https://github.com/VQAssessment/DOVER/issues/6), what should I do if I want to remove randomness in inference. The results obtained from each inference remain consistent.