PIAFusion_pytorch
PIAFusion_pytorch copied to clipboard
Question for the Dataloader
Hi,
Sorry to have another question for your implementation. In your Dataloader implementation, I notice that you alway return the whole images and feed them to the model for model training, however; in the paper, it seems that they split images to 64x64 image patches with stride 64. I would like to know whether my observation is correct or not.
Thanks, Li-Yun
训练的时候使用的是来自于作者提供的数据集文件,作者提供的分割是直接对图片做了预处理然后保存成h5文件,我这里是直接将h5文件中的数据又还原回了图片形式,因此不需要在Dataloader中加入分割图片的操作
您好!我在一个遥感图像网站上下载了训练集和验证集,我解压发现训练集里的Image和mask文件都是.h5文件,然后我按照网上的教程查看了Image文件里面的每个.h5都只有一个key是img,文件里包含14个波段的数据,想请问一下,您知道怎么将这种多波段的.h5文件转换成图片吗?