search

JingyunLiang / MANet

Official PyTorch code for Mutual Affine Network for Spatially Variant Kernel Estimation in Blind Image Super-Resolution (MANet, ICCV2021)
https://arxiv.org/abs/2108.05302
Apache License 2.0
169 stars 23 forks source link

some question about calculating psnr_kernel_y in stage1 #10

Open leeziqiang opened 3 years ago

leeziqiang commented 3 years ago

您好，请教个问题，在stage1阶段的validation和测试阶段，计算kernel的psnr_kernel_y 是选取了一个点的kernel来计算的，不用选取图像的所有的点的kernel吗，validation选的是visuals['KE'][5000, :, :]，而test阶段是visuals['KE'][300, :, :]，不是很理解，另外我的代码在stage1的时候在验证集上的psnr_kernel_y 不是很稳定，在53，54，55之间变化。

JingyunLiang commented 3 years ago

visuals['KE'][5000, :, :] is just for reference. We use LR image PSNR for evaluation. See Sec. 4.1 (performance evaluation) for more details.