Overview: The repository records a path of chasing faster ConvNet.
The repo is still under construction!
Edge-enhanced Feature Distillation Network for Efficient Super-Resolution
Yan Wang
Nankai University
Summary: 5th solution of Model Complexity in the NTIRE 2022 Challenge on Efficient Super-Resolution. Involoving the modification of convolution and network architecture.
- 🌟 Convolution: edge-ehanced reparameter block (EDBB) with a corresponding edge loss .
- 📦 Attention: original ESA.
- 📦 Backbone: backbone searched by network-level NAS.
Partial Feature Distillation Network for Efficient Super-Resolution
Yan Wang, Erlin Pan, Qixuan Cai, Xinan Dai
Nankai University, University of Electronic Science and Technology of China, Tianjin University
Summary: Winner of Overall Evaluation and 4th of Runtime in the NTIRE 2023 Challenge on Efficient Super-Resolution. Involoving the modification of convolution and network architecture.
- ⭐️ Convolution: integrating partial convolution and RRRB.
- 📦 Attention: efficient ESA.
- 📦 Backbone: ResNet-style backbone.
| Model | Runtime[ms] | Params[M] | Flops[G] | Acts[M] | GPU Mem[M] |
|---|---|---|---|---|---|
| RFDN | 35.54 | 0.433 | 27.10 | 112.03 | 788.13 |
| PFDN | 20.49 | 0.272 | 16.76 | 65.10 | 296.45 |
Lightening Partial Feature Distillation Network for Efficient Super-Resolution
Yan Wang, Yi Liu, Qing Wang, Gang Zhang, Liou Zhang, Shijie Zhao
Nankai University, ByteDance
Summary: 3rd of Overall Evaluation and 3rd of Runtime in the NTIRE 2024 Challenge on Efficient Super-Resolution. Involoving the modification of convolution, attention and network pruning.
- 📦 Convolution: RepMBConv in PlainUSR.
- 📦 Attention: LIA in PlainUSR.
- ⭐️ Backbone: ABPN-style backbone and block pruning.
To be updated.
PlainUSR: Chasing Faster ConvNet for Efficient Super-Resolution
Yan Wang, Yusen Li†, Gang Wang, Xiaoguang Liu
Nankai University
Summary: we present PlainUSR incorporating three pertinent modifications (convolution, attention, and backbone) to expedite ConvNet for efficient SR.
- 🌟 Convolution: Reparameterized MobileNetV3 Convolution (RepMBConv).
- ⭐️ Attention: Local Importance-based Attention (LIA).
- 🌟 Backbone: Plain U-Net.
To be updated.
To be updated.
We would thank BasicSR, ECBSR, DBB, ETDS, FasterNet, etc, for their enlightening work!
@inproceedings{wang2022edge,
title={Edge-enhanced Feature Distillation Network for Efficient Super-Resolution},
author={Wang, Yan},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
pages={777--785},
year={2022}
}
@article{wang2024plainusr,
title={PlainUSR: Chasing Faster ConvNet for Efficient Super-Resolution},
author={Wang, Yan and Li, Yusen and Wang, Gang and Liu, Xiaoguang},
journal={arXiv preprint arXiv:2409.13435},
year={2024}
}