READ ME File For 'Dataset for "GhostShiftAddNet: More Features from Energy-Efficient Operations"

Dataset DOI:  10.5258/SOTON/D2007

ReadMe Author: Jia Bi, University of Southampton [0000-0002-3773-3289]

This dataset supports the publication:
AUTHORS: Jia Bi, Jonathon Hare, Geoff V. Merrett 
TITLE: GhostShiftAddNet: More Features from Energy-Efficient Operations
CONFERENCE: The 32nd British Machine Vision Conference 2021
PAPER DOI:


This dataset contains:

---- Figure. 1 ---- 
An illustration of proposed GhostSA module for outputting the same number of features. SConv applies the number of bit-shift filters to generate intrinsic features from the input features. Then a modified DWSConv utilises bit-wise shift and addition operations to generate a series of ghost features.

---- Figure. 2 ----
This figure shows two version of GhostSA bottleneck when stride=1 and stride=2.

---- Table. 1 ----
Filename: ‘Table 1.xlsx’
Data: Experimental results illustrating comparison the performance of Ghost modules and GhostSA modules with different γ for compressing ResNet20 and VGG-16 on CIFAR10.

---- Figure.3 ----
Filename: ‘Figure 3.xlsx’
Data: Comparison of FLOPs and latency of state-of-the-art lightweight networks against accuracy on the CIFAR10 dataset.


---- Table. 2 ----
Filename: ‘Table 2.xlsx’
Data: Experimental results shows comparison of state-of-the-art networks over classification accuracy, the number of weights and FLOPs on the ImageNet dataset

---- Table.3 ----
Filename: ‘Table 3.xlsx’
Data: Experimental results shows GhostSA applied backbone models on Jetson Nano for CIFAR10 that is shown as G-backbones. All benchmarks are implemented by us

Date of data collection: Feb - May 2020

Information about geographic location of data collection: Southampton, UK

Licence: CC BY

Related projects: International Centre for Spatial Computation

Date that the file was created: Jun 2021