Kfashion Style classifier - Multi Label classification

Reference

• EfficientNet : https://arxiv.org/pdf/1905.11946.pdf
• Asymmetric Loss : https://arxiv.org/pdf/2009.14119.pdf

Model Description

EfficientNet Model Architecture	Activation Map

Asymmetric Loss Description

Formula	Comparison

Requirements

• python V # python version : 3.8.13
• dgl==0.9.1
• tqdm
• torch==1.9.1
• torchvision==0.10.1
• torchaudio==0.9.1
• torchtext==0.10.1
• dask
• partd
• pandas
• fsspec==0.3.3
• scipy
• sklearn

cmd running

The install cmd is:

conda create -n your_prjname python=3.8
conda activate your_prjname
cd {Repo Directory}
pip install -r requirements.txt

• your_prjname : Name of the virtual environment to create

If you want to proceed with the new training, adjust the parameters and set the directory and proceed with the command below.

The Training cmd is:

python3 style_train.py 

The testing cmd is:

python3 Inference_inspection_kfashiontest.py 

The inference cmd is:

python3 Inference.py 

Test Result

Testset Distribution

testset fashion category

• Model Performance Table

test performance

Model	Class Num	Testset Num	Top3 Recall
Global Convolutional Network	10	41,178	91.1%
EfficientNet	23	118,483	95.5%

Class	Number	Top3 Recall
preppy	1,218	89.8%
resort	29,757	94.0%
punk	389	91.9%
classic	14,809	91.4%
military	1,656	95.4%
sporty	6,710	94.4%
retro	3,512	93.5%
oriental	1,704	90.6%
country	13,792	93.1%
hiphop	1,254	88.0%
hippy	2,644	93.8%
avantgarde	1,576	89.8%
modern	31,242	93.6%
romantic	28,976	95.0%
manish	3,382	88.2%
genderless	6,003	92.8%
kitsch	2,858	90.6%
tomboy	3,921	88.7%
street	134,410	97.5%
feminine	34,652	93.9%
western	665	88.7%
sophisticated	11,960	91.9%
sexy	3,714	90.8%

• Example