Qiu Jueqin

Hi, I notice you implement multiple batch training by looping every single image within a batch, as can be seen [here](https://github.com/jytime/Mask_RCNN_Pytorch/blob/master/model.py#L1722). Why not use PyTorch built-in multiple batch training scheme?

Hi, could you provide the FLOPs or MAdds of the CSRNet, with comparison to HDRnet or other lightweight networks? Thanks.

How large was the loss after 160000 iterations for a typical Enhance task?

1

Hi, I tried to reproduce Enhance stage in my own datasets (autonomous city scenarios, not special), but after 160000 iterations, both `loss_c + loss_const` and `loss_s_1 + loss_s_2` are still...

A crucial bug when calculating discriminator loss?

1

Hi mtics, I notice you calculate discriminator loss [here](https://github.com/mtics/deep-photo-enhancer/blob/8f004fd5c6a4191d983baff985da76b254eb31eb/libs/compute.py#L304) as per Eq.12 in the paper, but this objective function should be maximized, instead of being minimizing. Maybe this is the...

对于中文句子中的解释性英文短语，应该使用中文括号还是西文括号？

1

是应该使用 『我们公开了目前最好的（State-of-the-Art）模型』 还是 『我们公开了目前最好的 (State-of-the-Art) 模型』 呢？

Hi, I encountered a stuck issue when training in DDP, and here is the reason I figured out: https://github.com/megvii-research/PCB/blob/066cd923360301acd097d0ce5db6ce996a0e8675/mmdet/models/losses/cross_entropy_pcb_loss.py#L281-L291 the if branch starting from line 284 may be entered for...