Imperial College London
Portrait Picture

DrEdwardJohns

Faculty of Engineering, Department of Computing

Senior Lecturer
 
 
 
//

Contact

 

e.johns Website

 
 
//

Location

 

365ACE ExtensionSouth Kensington Campus

//

Summary

 

Publications

Citation

BibTex format

@inproceedings{Liu:2020:10.1007/978-3-030-58610-2_39,
author = {Liu, S and Lin, Z and Wang, Y and Jianming, Z and Perazzi, F and Johns, E},
doi = {10.1007/978-3-030-58610-2_39},
pages = {661--677},
publisher = {Springer Verlag},
title = {Shape adaptor: a learnable resizing module},
url = {http://dx.doi.org/10.1007/978-3-030-58610-2_39},
year = {2020}
}
Download

RIS format (EndNote, RefMan)

TY  - CPAPER
AB  - We present a novel resizing module for neural networks: shape adaptor, a drop-in enhancement built on top of traditional resizing layers, such as pooling, bilinear sampling, and strided convolution. Whilst traditional resizing layers have fixed and deterministic reshaping factors, our module allows for a learnable reshaping factor. Our implementation enables shape adaptors to be trained end-to-end without any additional supervision, through which network architectures can be optimised for each individual task, in a fully automated way. We performed experiments across seven image classification datasets, and results show that by simply using a set of our shape adaptors instead of the original resizing layers, performance increases consistently over human-designed networks, across all datasets. Additionally, we show the effectiveness of shape adaptors on two other applications: network compression and transfer learning.
AU  - Liu,S
AU  - Lin,Z
AU  - Wang,Y
AU  - Jianming,Z
AU  - Perazzi,F
AU  - Johns,E
DO  - 10.1007/978-3-030-58610-2_39
EP  - 677
PB  - Springer Verlag
PY  - 2020///
SN  - 0302-9743
SP  - 661
TI  - Shape adaptor: a learnable resizing module
UR  - http://dx.doi.org/10.1007/978-3-030-58610-2_39
UR  - http://hdl.handle.net/10044/1/83622
ER  -
Download