Results of data training

[TQG1997]

I tried to train these data, but failed to generate an MP4 file. I changed the format to GIF and started training successfully. According to up, are there more than 70000 offensive rounds? I have been training with V100 for a week and only trained 20000 rounds.

I found that the GIF generated by training does not include the picture of goals. Is it until the players shoot?

[chychen]

thanks for being interested in our work.

• about the mp4 generation failure, I think it might relate to the development environment, it was tested 4 years ago.
• about training time, it should be finished in a few hours with a single V100 GPU, something wrong there, however, the project had been no longer maintained, I could not give you any helpful comments.
• about data collection, we randomly trim the sequences into fixed lengths for training.

btw, I think you might also be interested in another work based on imitation learning and reinforcement learning, please checkout out https://github.com/chychen/RL_strategies. (However, again it had been no longer maintained since 3 years ago.)

thanks again, and good luck with your research.

[TQG1997]

Thank you very much. You can still reply to my questions！ 

 

------------------ 原始邮件 ------------------ 发件人: "Jay @.>; 发送时间: 2022年6月28日(星期二) 下午3:33 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [chychen/BasketballGAN] Results of data training (Issue #5)

感谢您对我们的工作感兴趣。

关于 mp 4 一代的失败，我觉得可能和开发环境有关， 4 年前就测试过了。

关于训练时间，它应该在几个小时内完成与一个单一的 V100 GPU ，有些错误那里，然而，该项目已经不再维护。

关于数据汇集，我们随机修剪成固定长度的序列进行训练。

btw ，我想你可能也会对另一个基于模仿学习和强化学习的作品感兴趣，请去看看https://github.com/chychen/RL_strategies.（然而，再次，它已不再维持自 3 年前。

再次感谢，祝你研究顺利。

— 直接回复此邮件，在 GitHub 上查看或取消订阅. 您收到这个消息是因为您创建了这个线程。留言 ID @.***与>

[TQG1997]

According to this picture, how long did it take you to reach 200K epochs

[chychen]

Sorry, I can't remember, it should be less than 6 hours in single 1080Ti or 2080Ti.

[TQG1997]

我想拜读一下您的《生成對抗網路應用於籃球防守戰術生成》这篇文章，能否将PDF文件发到我的邮箱 @.***

 

------------------ 原始邮件 ------------------ 发件人: "chychen/BasketballGAN" @.>; 发送时间: 2022年7月4日(星期一) 中午11:55 @.>; @.**@.>; 主题: Re: [chychen/BasketballGAN] Results of data training (Issue #5)

Sorry, I can't remember, it should be less than 6 hours in single 1080Ti or 2080Ti.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

[chychen]

@TQG1997 FYI https://github.com/chychen/bball_defensive_strategies_generation https://github.com/chychen/RL_strategies https://github.com/chychen/BasketballGAN