Skip to content

boostcampaitech7/level2-cv-semanticsegmentation-cv-14-lv3

Repository files navigation

Hand Bone Image Segmentation

📖 Overview

  • Duration : 2024.11.13 10:00 ~ 2024.11.28 19:00
  • 네이버 커넥트 재단 및 Upstage에서 주관하는 비공개 대회
  • X-ray Hand bone 이미지를 이용해 Segmentation Task를 수행하는 모델을 개발하는 대회
  • 하나의 이미지당 29개의 class를 가지고 있고 왼손, 오른손 동일한 양의 이미지가 존재

🧑‍💻 Contributors

김태한 문채원 서동환 윤남규 이재훈 장지우






📝 Wrap up Report

Data-Centric report
  • Dice
image

🔧 Tools

  • 🧑‍💻 Programming : GitHub, VScode
  • 👥 Communication : GitHub, Notion, Slack
  • 🧐 Monitoring and report : WandB
  • 💄 Visualization : Streamlit, Gradio, WandB
  • 🧱 Deployment : Docker

📦 Folder Structure

📦level2-cv-semanticsegmentation-cv-14-lv3
 ┣ 📂archive
 ┃ ┣ 📜gpu_trainer.py
 ┃ ┣ 📜trainer_hook.py
 ┃ ┣ 📜train_hook.py
 ┃ ┗ 📜tta_inference.py
 ┣ 📂instance_seg
 ┃ ┣ 📜convert_dataset.py
 ┃ ┗ 📜yolo_train.py
 ┣ 📂model
 ┃ ┣ 📂duck_net
 ┃ ┣ 📂u3_effnet
 ┃ ┣ 📂u3_maxvit
 ┃ ┗ 📂u3_resnet
 ┣ 📂tools
 ┃ ┣ 📂2_stage
 ┃ ┃ ┗ 📜ROI_Extraction.py
 ┃ ┣ 📂ensemble
 ┃ ┃ ┣ 📜2class_ensemble.py
 ┃ ┃ ┣ 📜fusion_new.py
 ┃ ┃ ┣ 📜hard_ensemble.py
 ┃ ┃ ┣ 📜merge_wrist.py
 ┃ ┃ ┣ 📜soft_ensemble.py
 ┃ ┃ ┣ 📜soft_voting_setting.yaml
 ┃ ┃ ┗ 📜weight_ensemble.py
 ┃ ┣ 📂streamlit
 ┃ ┃ ┣ 📜aug_vis.py
 ┃ ┃ ┣ 📜vis.py
 ┃ ┃ ┣ 📜visualize.py
 ┃ ┃ ┗ 📜vis_test.py
 ┃ ┗ 📜csv_merger.py
 ┣ 📂utils
 ┃ ┣ 📜weight_init.py
 ┃ ┗ 📜__init__.py
 ┣ 📜dataset.py
 ┣ 📜functions.py
 ┣ 📜inference.py
 ┣ 📜loss.py
 ┣ 📜requirements.txt
 ┣ 📜sweep_config.yaml
 ┣ 📜train.py
 ┗ 📜trainer.py

Used Architecture

Semantic Segmentation Models

  • U-Net 계열 : UNet, UNet++, UNet3+, DuckNet, swinUNETR
  • Yolo 계열 : YOLOv8x-seg, YOLOv11x-seg
  • DeepLab 계열 : DeepLabV3, DeepLabV3+

Backbones(Encoder)

  • ResNet 계열 : ResNet, ResNeXt, ResNeSt,
  • EfficientNet 계열 : B4, B5, B6, B7, timm-b7, V2-L
  • ViT 계열 : MiT, MaxViT
  • ETC : HRNet, Swin-T, DUCK-Net

LB Score

  • Public Score
image
  • Private Score
image

About

level2-cv-semanticsegmentation-cv-14-lv3 created by GitHub Classroom

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published