Few-Shot Fine-Grained Image Classification Based on Multiscale Joint Distribution

Table 1. Information of experimental datasets
View table
Table 1. Information of experimental datasets
Dataset Number of categories
Total Training set Validation set Test set
CUB-200-2011 200 100 50 50
Stanford Cars 196 130 17 49

Table 2. Classification accuracy on CUB-200-2011

View table

Table 2. Classification accuracy on CUB-200-2011

Method	Accuracy /%
Method	5-way 1-shot	5-way 5-shot
MAML^［22］	68.42±1.07	83.47±0.62
Baseline++^［23］	67.02±0.90	83.58±0.54
MatchNet^［24］	71.87±0.85	85.08±0.57
ProtoNet^［12］	80.90±0.43	89.81±0.23
FRN^［15］	82.55±0.19	92.98±0.10
ADM^［25］	79.31±0.43	90.69±0.21
CovNet^［26］	80.76±0.42	92.05±0.20
Meta DeepBDC^［19］	85.55±0.40	93.82±0.17
FS-FGIC^［27］	86.14±0.18	95.08±0.09
MS-JDFM（ours）	87.22±0.20	96.51±0.18

Table 3. Classification accuracy on Stanford Cars

View table

Table 3. Classification accuracy on Stanford Cars

Method	Accuracy /%
Method	5-way 1-shot	5-way 5-shot
DN4^［28］	61.51±0.44	89.60±0.44
MsKPRN^［29］	76.64±0.84	89.88±0.46
BSNet^［30］	67.48±0.22	86.88±0.50
SAML^［31］	75.74±0.20	88.73±0.49
FRN^［15］	84.01±0.17	93.75±0.07
CTX^［32］	85.03±0.19	92.63±0.11
ProtoNet^［12］	83.46±0.19	92.08±0.08
Meta DeepBDC^［19］	88.26±0.26	94.13±0.13
FS-FGIC^［27］	88.96±0.37	95.16±0.20
MS-JDFM（ours）	90.65±0.12	97.78±0.37

Table 4. Comparison of model parameter quantities
View table
Table 4. Comparison of model parameter quantities
Method Number of parameters /10⁶ Stanford Cars accuracy /%
DeepBDC^［19］ 12.8 94.13±0.13
FS-FGIC^［27］ 14.3 96.12±0.17
MS-JDFM（ours） 13.5 97.78±0.37

Table 5. Ablation experiment results on CUB-200-2011
View table
Table 5. Ablation experiment results on CUB-200-2011
Method Accuracy /%
5-way 1-shot 5-way 5-shot
DeepBDC 85.55±0.40 93.82±0.17
A 85.96±0.42 94.35±0.20
B 86.85±0.40 96.12±0.17
MS-JDFM（ours） 87.22±0.20 96.51±0.18

Table 6. Ablation experiment results on Stanford Cars
View table
Table 6. Ablation experiment results on Stanford Cars
Method Accuracy /%
5-way 1-shot 5-way 5-shot
DeepBDC 88.26±0.26 94.13±0.13
A 88.86±0.19 94.88±0.08
B 90.22±0.26 97.34±0.13
MS-JDFM（ours） 90.65±0.12 97.78±0.37

Tools

Get Citation

Copy Citation Text

Shudong Liu, Zeyu Hao, Honghui Wang, Jia Cong, Boyu Gu. Few-Shot Fine-Grained Image Classification Based on Multiscale Joint Distribution[J]. Laser & Optoelectronics Progress, 2025, 62(14): 1437002

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites