zcong_huang

> Hi kwea123, thanks for your attention, previously, I evaluate nerf with your implementation, I think there are two main differences here: 1) we trained the scenes with 16 training...

I tried not to use this part，and train the model . It didn't matter and the result is similar。

>  > Is it normal? > The last three output maps are always all black after many epoches. I met the same problem as you, Have you solved it?...

I think i have found the solution. Change the kernel size of [upsample layer5](https://github.com/meteorshowers/RCF-pytorch/blob/master/models.py#L141) from 32 to 16, the problem will be solved.

> You're right! > I'll fix it. 'tuple' object can not be modified. Your code just get an generator.

> 你好，我的训练速度也很慢，显卡跟您一样，您最后找到速度慢的原因了吗 你好，我最近想跑通一下试试，结果发现训练特别慢，一个iteration需要好几秒，而且显存根本加载不上去，但是又确实在跑。很奇怪这种现象，不知道你是否也一样，或者你现在找到解决方法了嘛？

Hi Henri , Thanks a lot ! I meet the same problem as Jeremy's. `rosrun dvs_renderer dvs_renderer events:=/cam0/events` After running this, it got stuck. I don't know if it's running...

不好意思，有个问题漏掉了 5. 对于Vsource与Vreference，论文中您给出的公式是直接做差`Vi;d;h;w = Vi;d;h;w - V0;d;h;w;`，但是在代码上，是求[平方差](https://github.com/yhw-yhw/PVAMVSNet/blob/master/models/vamvsnet.py#L452)`warped_volume = (warped_volume - ref_volume).pow_(2)` 请问两者最后的效果有区别吗？ 再次感谢您！