Readme

Description

This repo was generated for participating in the upstage contest. I got 79.9524% Accuracy and 0.7528 F1-score in public LB, 79.1429% Accuracy and 0.7333 F1-score in final LB.

• Goal : Image classification
  We need a system that automatically identifies whether this person is wearing a mask or not, and whether he or she is wearing it correctly, just by using the image of the person's face shown on the camera.

• Transfer-learning : Custom model was finetuned-model with pre-trained resnet50 model. Experiments were conducted with changing data augmentation, loss, optimizer, lr scheduler etc.

{   
    "seed": 42,
    "epochs": 20,
    "resize": [224,224],
    "batch_size": 64,
    "valid_batch_size": 64,
    "model": "resnet50(pretrained=True)",
    "optimizer": "Adam",
        "scheduler" : "StepLR",
    "gamma" : 0.1,
    "weight_decay": 0.0005,
    "lr": 0.0001,
    "val_ratio": 0.9,
    "criterion": "cross_entropy",
    "lr_decay_step": 7,
}

Installation

pip install -r requirements.txt at ./code dir.

torch==1.6.0
torchvision==0.7.0
tensorboard==2.4.1
pandas==1.1.5
opencv-python==4.5.1.48
scikit-learn~=0.24.1
matplotlib==3.2.1

Contraints

• In this repo, input directory doesn't exist!
  • train.csv and submission.csv was not uploaded. But the column info must contains ImageID, ans header.

Improvements

• Things to improve.
• Will be updated.