CSCN_ICCV.md

Short introduction

Sparse coding and deep learning

Main contributions

• combine sparse coding and deep learning
• using network cascading
• conduct a subjective evaluation

Architecture

Overall

Cascade with mnulti-scale objectives

Loss

• MSE
• Loss function:

Training strategy

Experiments

• Dataset: 91
• Evaluation metric: PSNR
• Patchsie: input 56 × 56 output 44 X 44
• Dict: sparse coding, LR: 9X9 HR: 5 X 5
• SGD

Final summary

Pros:

• Combine sparse coding with deep learning as guidiance

Cons:

Tips:

• Multi-scale objectives