Namhyuk Ahn

In the RealSR dataset, we have observed that early stopping is crucial for the best performance. The overfitting seems to worsen when the network is bigger (e.g. EDSR, RCAN) and...

Train an SR model with a large patch size improves the performance, but it also drastically increases the training time and the memory consumption so that most of the methods...

First of all, in our internal analysis, we have found that the nn-based upsample is better than the bicubic (although it's marginal). Unlike the early-upsample SR methods which mostly upsample...

I think that the latest torchtext remove train argument of the process function. Check this code [field.py#L225](https://github.com/pytorch/text/blob/master/torchtext/data/field.py#L225). Erase `train=True` makes it work.

Hi. Did you run the code on Window? My code is only tested on the Linux and I never have encountered such issue. And I guess that this stackoverflow [question](https://stackoverflow.com/questions/49386121/python-h5py-file-read-oserror-unable-to-open-file-bad-superblock-version-numb)...

hmm... Have you evaluated the PSNR of the provided result as well? And are other cases are matched to the paper? Unfortunately, I can't check this issue since I don't...

Hi. 1) Only CARN-M uses residual-e block (group convs) as in the code. 2) In the section 4.4, PSNR of CARN is only calculated for the ablation study so that...

As you say, since they didn't publish the result images, we only record the performance result that borrowed from their paper. For the qualitative comparison like Fig 6, we put...

Hi, Could you let me know your system setting? Mine is Python **3.6** and PyTorch 0.3 and checked it works well. Perhaps, `u[-1]@self.c ...` could be a problem. `@` symbol...

Hi. Is the training time cumulatively increased by 7s in every 1k steps? I haven't plotted the wall-time graph, so I haven't known that issue and neither not sure how...