boostcampaitech3 / level2-semantic-segmentation-level2-cv-16

[부스트캠프 AI Tech 3기 / CV-16] 재활용 품목 분류를 위한 Semantic Segmentation (22.04.25 - 22.05.13)
1 stars 4 forks source link

Team Medic(CV-16)

Project Overview

Project Period 2022.04.25 ~ 2022.05.12

🔎 재활용 품목 분류를 위한 Segmentation

image

😎 Members

권순호 서다빈 서예현 이상윤 전경민
Github Github Github Github Github

🌏 Contribution

❓Problem Definition

image

🚨 Competition Rules

💾 Datasets

image


image

💾 Annotations

💻 Development Environment

📁 Project Structure

level2-semantic-segmentation-level2-cv-16
├─ code
│   ├─ baseline_fcn_resnet50.ipynb
│   ├─ class_dict.csv
│   ├─ requirements.txt
│   ├─ utils.py
│   ├─ recycledataset.py
|   ├─ train_unetthreeplus.py
|   ├─ unetthreeplus.py
|   ├─ copy_pasteaug.py
│   ├─ saved
│   └─ submission
│       └─ sample_submission.csv
└─ data
    ├─ test.json
    ├─ train.json
    ├─ train_all.json
    ├─ val.json
    ├─ batch_01_vt
    │   ├─ 0002.jpg
    │   ├─ ...
    │   └─ 0005.jpg
    ├─ batch_02_vt
    │   ├─ 0001.jpg
    │   ├─ ...
    │   └─ 0003.jpg
    └─ batch_03
        ├─ 0001.jpg
        ├─ ...
        └─ 0003.jpg

👨‍🏫 Evaluation Methods

image

💯 Final Score

image

👀 How to Start

git clone https://github.com/boostcampaitech3/level2-semantic-segmentation-level2-cv-16.git
cd level2-semantic-segmentation-level2-cv-16.git
pip install -r requirements.txt
python stratified_kfold/kfold.py
python hard_vote_ensemble/ensemble.py
  1. mmsegmentation ( Link )
    • Installing prerequisites (without Albumentations)
pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu110/torch1.7/index.html
pip install mmsegmentation
cd mmsegmentation
conda create -n mmseg python=3.8
conda activate mmseg
pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu110/torch1.7/index.html
conda install pytorch==1.7.1 torchvision==0.8.2 torchaudio==0.7.2 cudatoolkit=11.0 -c pytorch
conda install -c conda-forge albumentations
conda install -c conda-forge wandb
pip install -e .
cd mmsegmentation
python tools/train.py <<directory_of_config_file>>
  1. segmentation_modules_pytorch
python train_unetthreeplus.py --data_path <<parentdir_path_of_datasets>>\
--train_path train.json --valid_path val.json --test_path test.json \
--encodername < ex) tu-efficientnet_b4 >

📄 Experiments & Submission Report