SplitSR

Unofficial implementation of SplitSR: An End-to-End Approach to Super-Resolution on Mobile Devices

Keys from the Paper

• Split convolution splits input by alpha ratio along depth channel.
• The conv-processed part is concatenated at the end.
• By the second key point, every channel would be processed after 1/α blocks.
• The theoretical computation reduction that can be obtained by using SplitSR is 𝛼^2, where 𝛼 ∈ (0, 1]
• The architecture is very much similar to RCAN's, by replacing channelwise attention blocks with split convolutions.
• Many proposed details are ambiguous. We've to guess.

Config

• 𝛼 = 0.250
• Groups = 6, Blocks = 6
• Hybrid Index = 3
• Loss - L1
• Base LR - 1e-4
• LR Decay - 2.0 every 2 × 10^2
• Adam, 𝛽1 = 0.9, 𝛽2 = 0.999
• 𝜖 = 1e−7
• Steps = 6 × 10^5

Progress

• Splitted Convolution Block is done.
• Residual Block is done.
• Mean shift layer is done.
• Beta version of model is ready.