ksm0517 / semantic-segmentation

semantic-segmentation-level2-cv-07 created by GitHub Classroom

Level2 P-stage Semantic Segmentation

💡 Team: 컴퓨터구조

Project Overview

• Trash semantic segmentation
• Input: 512 x 512 Image
  • Train: 2617 images
  • Validation: 655 images
• Output: Category classification for each pixel
  • Class(11): Background, General trash, Paper, Paper pack, Metal, Glass, Plastic, Styrofoam, Plastic bag, Battery, Clothing

Archive contents

image-classification-level1-02/
├── input/data/
│   ├── batch_01_vt/
│   ├── batch_02_vt/
│   ├── batch_03/
│   ├── train.json
│   ├── val.json
│   ├── train_all.json
│   └── test.json
├── mmsegmentation/
│   ├── configs/
│   ├── mmdetection library folders and files ...
│   ├── train.py
│   └── inference.py
└── util/

• input/data/ : download from https://stages.ai/

get start

train & inference

cd mmsegmentation

python train.py
python inference.py

visualize

cd util

converting visualize_CSVs py to ipynb

jupyter notebook visualize_CSVs.ipynb

# set result csv files
csv_names = ['Unet_resnet50.csv', 'deeplabv3_resnet50_31epoch.csv']

Requirements

• Ubuntu 18.04.5
• Python 3.8.5
• pytorch 1.7.1
• torchvision 0.8.2

Install packages : pip install -r requirements.txt

Hardware

• CPU: 8 x Intel(R) Xeon(R) Gold 5220 CPU
• GPU: V100
• RAM: 88GB

Contributors

Name @github
고재욱 @고재욱
김성민 @ksm0517
박지민 @박지민
박진형 @ppjh8263
심세령 @Seryoung Shim
윤하정 @Yoon Hajung

Data Citation
네이버 커넥트재단 - 재활용 쓰레기 데이터셋 / CC BY 2.0