Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Wait on every block #82

Open
james-s-w-clark opened this issue Jan 13, 2018 · 3 comments
Open

Wait on every block #82

james-s-w-clark opened this issue Jan 13, 2018 · 3 comments

Comments

@james-s-w-clark
Copy link

james-s-w-clark commented Jan 13, 2018

(From watching the demo video) It looks like on every block, the avatar waits for long enough to get points from a shop/music/rubik's cube/shower tile. However, it only has to wait if it is on one of these blocks.

看起来每次等拿点数,可是一般不在 音乐/超市/魔方的地方。

Could be easy (get colour of place) to see if it's worth waiting. Maybe easy to build this into the training too.

可能是个很容易的东西 (可以看看地方的颜色),决定如果应该等拿点数。

@Prinsphield
Copy link
Owner

Prinsphield commented Jan 13, 2018

其实每次等待的时间都是1.5s,这个是为了保证小人落稳,没有特效的干扰。

中间找位置还需要花一点时间,我们的三种方法花的时间大约是
Fast search: 0.004s
Multi-scale search: 1.9s
CNN search (CPU): 1.9s

我对比了其他几个项目,在保证准确度的情况下,我们的方法还是比较快的。算上小人中间跳的过程花的时间,最坏情况都是4s跳一次,平均情况会远好于这个,因为我们基本上都跳在中心,这时候用fast search,大概2s多一跳。

这些时间足够让小人把该有的特效都拿到。

@Prinsphield
Copy link
Owner

Prinsphield commented Jan 13, 2018

当然,其实我们的cnn模型完全可以用一个更小的模型学好,甚至也可以end to end训练,把时间消耗做得更低,只不过做这个项目指是个业余爱好,没有再花更多的时间去调优。

另外我们一开始选择较为慢的TensorFlow而不是更快的pytorch,是因为我们考虑到很多用户是windows系统,而TensorFlow对windows有官方的支持。

@zkailinzhang
Copy link

@Prinsphield,请问端到端的网络怎么做,能简单说下思路吗,正在用你采集的数据训练,和这个cnn有什么训练区别呢

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants