mango

i meet the same problem. i change '--gpus' , --gpus = 3, but only one gpu works，same mistakes

@JingyunLiang Thank you for your answer.if i want to get x2 result, what is the maximum input size supported?

@Xiangtaokong 谢谢您的回答，不过论文中描述patch_size=32，您说的图像块128*128步长16是什么意思呢？分类的时候图像块的大小为128*128吗？且相邻图像块之间有重复的区域？

@Xiangtaokong 是否可以不重合分割图像块呢？

@Xiangtaokong 麻烦请教您几个问题： 1、论文中指出模型包含不同通道数的三个分支，分支数是如何确定的？ 2、三个分支是分别训练并得到三个预训练模型吗？测试的时候根据分类结果调用网络分支时，是加载对应的预训练模型进行推理吗？ 3、训练过程是先进行train a single branch，后进行train ClassSR吗？

@LZleejean How did you deal with it? thanks.

@Hans1984 The pixel value of the .hdr format image is large in the dataset .When saved, the value range is [0,1]. Is the image information obtained not lost ?

> 请问cuda版本必须是10.0吗？我的是11.0，修改make.sh后，编译报错 > > make.sh: line 1: i#!/usr/bin/env: No such file or directory running build running build_ext building '_ext' extension Emitting ninja build file /mnt/VSR/TMNet/models/modules/DCNv2/build/temp.linux-x86_64-3.6/build.ninja... Compiling objects... Allowing ninja to...

> > > Hello, > > [MSU Graphics & Media Lab Video Group](https://videoprocessing.ai/) has recently launched two new Super-Resolution Benchmarks. > > > > * [Video Upscalers Benchmark: Quality Enhancement](https://videoprocessing.ai/benchmarks/video-upscalers.html)...

This is happening on my RTX6000 as well.if use cpu,very slow.