SR_Framework

A generic super-resolution framework which implements the following networks (Updating...)

• DLSR [Differentiable architecture search]
• RFDN [AIM20 Champion
• LatticeNet [ECCV2020]
• IMDN [ACM MM2019]
• SRFBN [CVPR2019]
• IDN [CVPR2018]
• CARN [ECCV2018]
• RCAN [ECCV2018]
• MemNet [ICCV2017]
• EDSR [CVPR2017]
• DRRN [CVPR2017]
• LapSRN [CVPR2017]
• DRCN [CVPR2016]
• VDSR [CVPR2016]
• FSRCNN [ECCV 2016]

Implement some useful functions for article figures. Like the following:

• 1. generate_best: Automatically compare your method with other methods and visualize the best patches.

• 2. Frequency_analysis: Convert an image to 1-D spectral densities.

• 3. relation: Explore relations in fuse stage.(eg. torch.cat([t1, t2, t3, c4], dim=1) and then fuse them with 1x1 convolution)

• 4. feature_map: Visualize feature map.(average feature maps along channel axis)