Multi-Label-Image-Classification

Features

Code Decoupling: Decouple all data loading, network model construction, model training and validation

Rich Content: Providing rich evaluation indicators and functional functions

Function Functionality

• checkpoints: stores the weights of the trained model;
• datasets: Store datasets. And partition the dataset;
• logs: Stores training logs. Including losses and accuracy during training and validation;
• option. py: stores all the parameters required for the entire project;
• utils.py: Stores various functions. Including model saving, model loading, and loss function, etc;
• split_data. py: Divide the dataset;
• model. py: Building a neural network model;
• train.py: Train the model;
• predict. py: Evaluate the training model;
• model_transfer. py: Transfer .pth model to .onnx model.

Requirements

Required:

matplotlib==3.8.3 numpy==1.26.4 Pillow==9.5.0 Pillow==10.3.0 scikit_learn==1.5.0 torch==2.2.1 torchvision==0.17.1 tqdm==4.66.2

You can install these dependencies via pip:

pip install -r requirements.txt

Usage

1. Prepare Dataset

Prepare your dataset and place it according to the following requirements:

datasets
   images
       styles.csv

If you need to conduct code validity testing first, you can use Fashion-Product-Images-Small first.

2. Split Your Dataset

Run the following script to obtain the partitioned dataset, but you need to pay attention to modifying some paths:

python split_data.py

Then you can obtain the following data structure:

datasets
   images
       styles.csv
       train.csv
       val.csv

3. Modify Network

How many categories do you need to predict? You need to add a few category headers in model.py.

4. Train Your Dataset

Run the following script to train your dataset and output various parameters during the training and validation processes:

python train.py

5. Evaluate Your Model

Running the following code can evaluate the accuracy of your model. You can choose to evaluate the test dataset or individual images:

python predict.py

License

This project is licensed under the Apache 2.0 license. For detailed information, please refer to the LICENSE file.

Acknowledgement

Kaggle Dataset: Fashion-Product-Images-Small

CSDN: 王乐予-CSDN博客