yu45020

@Auth0rM0rgan I add the most recent model [checkpoint and ways to resume it](https://github.com/yu45020/Waifu2x/tree/master/model_check_points/CRAN_V2) for further training. Hope it helps.

My codes may seem messy >.< For the most recent model training, I use sqlite to store pre-processed images. The scaling step is defined at https://github.com/yu45020/Waifu2x/blob/3b8713d571a9a704207f8833c07a5533c0c0279b/Dataloader.py#L87 ``` def get_img_patches(self, img_file):...

That seems a nice idea, and it will be a good chance for me to learn something new in programming. By the way, this repo was created as a practice...

The new model is trained in fp16, but PyTorch doesn't support fp32 in CPU. You may still use it in CPU only for inference by changing this line in Nvidia...

The training script is merely a demo, and I don't have the skills and time to develop all purpose training script. Indeed, this repo was initially used as a practice...

I use around 4k images and cut them into around 300k small patches.

关于空洞卷，我一开始的想法是扩大感受野，然后接上连续的正常卷，把这两部分合起来再重复。 扩大了的感受野就可以被下一层补充细节。 比如像这样 ```python block = nn.Sequential(nn.Conv2D(10,10,dilation=2, kernel=3, padding=2), nn.Conv2D(10,10,dilation=1, kernel=3, padding=1), nn.Conv2D(10,10,dilation=1, kernel=3, padding=1) ``` 或者做成residual block，或者不同的顺序组合，又或者不同dilation rate 组合， 可惜都没效果 >

有意思。 最初的选择是从 [Loss Functions for Image Restoration with NeuralNetworks](https://arxiv.org/pdf/1511.08861.pdf) 。L1貌似已经够好了，就没多想。下一步选loss也是个不错的实验方案。

I have the same environment, at least for the packages listed above. I first use virtual environment and get the same error. Then I switch back the original environment and...

[Anacondas](https://anaconda.org/anaconda/python) contains lots of common packages with Python. I use it to build exe, and I always get it work, thought the size is really LARGE ( usually include pyqt,...