still found bad pixeles in movie subtitles area after retrain

[zhu2bowen]

First of all, thank you for your outstanding work! I would like to use the model to infer images from moives and TV shows. After adding captions to the dataset, I retrained the model. But I still get bad imagesas follows:

Can you give me some guidance?

[wyhhhhhhhh]

您好，可以问一下您重新训练这个模型用了什么样的显卡，花费多长时间吗？