Intra-inter channel attention for few-shot classification

Table 1. Structural parameters of fully connected network
View table
View in Article
Table 1. Structural parameters of fully connected network
层名称输出尺寸
输入 $N_{p r e} \times C$
全连接层 $N_{p r e} \times C / 2$
批归一化 $N_{p r e} \times C / 2$
ReLU $N_{p r e} \times C / 2$
全连接层 $N_{p r e} \times C$
Sigmoid $N_{p r e} \times C$
转换层 $N_{p r e} \times C$

Table 2. Parameter setting of convolutional neural network
View table
View in Article
Table 2. Parameter setting of convolutional neural network
模块名称
卷积层
$c_{1} : 64, c_{2} : 64, c_{3} : 64, c_{4} : 64$
$k_{1} : 3, k_{2} : 3, k_{3} : 3, k_{4} : 3$
批归一化 $c_{1} : 64, c_{2} : 64, c_{3} : 64, c_{4} : 64$
最大池化
$s_{1} : (2,2), s_{2} : (2,2),$
$s_{3} : (2,2), s_{4} : (2,2)$

Table 3. Parameter configuration of few-shot classification task in meta-train stage
View table
View in Article
Table 3. Parameter configuration of few-shot classification task in meta-train stage
N-way K-shot 元训练阶段
way shot query
5-way 5-shot 20 5 15
5-way 1-shot 30 1 15

Table 4. Experimental results for ICAFSC updating ICAM parameters in different training stages on MiniImagenet
View table
View in Article
Table 4. Experimental results for ICAFSC updating ICAM parameters in different training stages on MiniImagenet
模型准确率/%
5-way 1-shot 5-way 5-shot
ProtoNet 49.62±0.78 67.96±0.62
ICAFSC-pre 51.55±0.75 69.11±0.65
ICAFSC-tra 49.76±0.77 68.14±0.60
ICAFSC-pre+tra 50.70±0.75 68.87±0.63

Table 5. Experimental results for ICAFSC under different classification task configurations in pre-training phase on MiniImagenet

View table

View in Article

Table 5. Experimental results for ICAFSC under different classification task configurations in pre-training phase on MiniImagenet

模型	预训练任务			准确率/%
模型	Way	Shot	Query	5-way 1-shot	5-way 5-shot
ProtoNet	-	-	-	49.62±0.78	67.96±0.62
ICAFSC	5	100	15	51.55±0.75	69.11±0.65
ICAFSC	10	100	15	50.30±0.77	68.79±0.60
ICAFSC	20	100	15	49.97±0.79	68.47±0.66

Table 6. Classification scores of different methods with Conv4-64 as backbone on MiniImagenet dataset

View table

View in Article

Table 6. Classification scores of different methods with Conv4-64 as backbone on MiniImagenet dataset

模型	准确率/%
模型	5-way 1-shot	5-way 5-shot
ProtoNet^［8］	49.62±0.78	67.96±0.62
MatchNet^［14］	43.40±0.78	51.09±0.71
MAML^［16］	48.07±1.75	63.15±0.91
P-RelationNet^［17］	49.54±0.82	68.34±0.70
BOIL^［19］	49.61±0.16	66.45±0.37
D2D2^［20］	51.20±0.60	68.80±0.10
HyperProto^［21］	50.21±0.31	66.48±0.71
DAM^［22］	53.80±0.19	69.31±0.16
CSS^［23］	50.85±0.84	68.08±0.73
HyperShot^［23］	52.42±0.46	68.78±0.29
OVE^［24］	48.00±0.24	67.14±0.23
IMP^［25］	49.60±0.80	68.10±0.80
ICAFSC	51.55±0.75	69.11±0.65

Tools

Get Citation

Copy Citation Text

Liping YANG, Tianyang ZHANG, Yuyang WANG, Xiaohua GU. Intra-inter channel attention for few-shot classification[J]. Optics and Precision Engineering, 2023, 31(21): 3145

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category:

Received: Apr. 6, 2023

Accepted: --

Published Online: Jan. 5, 2024

The Author Email: YANG Liping (yanglp@cqu.edu.cn)

DOI:10.37188/OPE.20233121.3145

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Structural parameters of fully connected network

Table 1. Structural parameters of fully connected network

Table 2. Parameter setting of convolutional neural network

Table 2. Parameter setting of convolutional neural network

Table 3. Parameter configuration of few-shot classification task in meta-train stage

Table 3. Parameter configuration of few-shot classification task in meta-train stage

Table 4. Experimental results for ICAFSC updating ICAM parameters in different training stages on MiniImagenet

Table 4. Experimental results for ICAFSC updating ICAM parameters in different training stages on MiniImagenet

Table 5. Experimental results for ICAFSC under different classification task configurations in pre-training phase on MiniImagenet

Table 5. Experimental results for ICAFSC under different classification task configurations in pre-training phase on MiniImagenet

Table 6. Classification scores of different methods with Conv4-64 as backbone on MiniImagenet dataset

Table 6. Classification scores of different methods with Conv4-64 as backbone on MiniImagenet dataset

微信扫一扫：分享