DiffusionDet 뿌시기 (Deep Dive)

정확한 디버깅을 위해서 먼저 프로그램 (코드) 를 돌려보기로 한다. 

참고 : https://github.com/ShoufaChen/DiffusionDet

 

0. 도커 세팅 

# 베이스 이미지
FROM pytorch/pytorch:2.4.0-cuda12.4-cudnn9-devel

# 필수 패키지 설치
RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y \
    git \
    nano \
    wget \
    libgl1-mesa-glx \
    libglib2.0-0 \
    libxcb-xinerama0-dev \
 && rm -rf /var/lib/apt/lists/*

 

 

1. 깃 클론

git clone https://github.com/ShoufaChen/DiffusionDet

 

 

2. requirements 세팅하기

2.1. detectron2 

python -m pip install 'git+https://github.com/facebookresearch/detectron2.git'

 

3. 데이터셋 준비 (symbolic link)

mkdir -p datasets/coco
ln -s /path_to_coco_dataset/annotations datasets/coco/annotations
ln -s /path_to_coco_dataset/train2017 datasets/coco/train2017
ln -s /path_to_coco_dataset/val2017 datasets/coco/val2017

ln -s /usr/src/data/coco/annotations datasets/coco/annotations
ln -s /usr/src/data/coco/images/train2017 datasets/coco/train2017
ln -s /usr/src/data/coco/images/val2017 datasets/coco/val2017

 

 

4. pretrained-model 준비

mkdir models
cd models
# ResNet-101
wget https://github.com/ShoufaChen/DiffusionDet/releases/download/v0.1/torchvision-R-101.pkl

# Swin-Base
wget https://github.com/ShoufaChen/DiffusionDet/releases/download/v0.1/swin_base_patch4_window7_224_22k.pkl

cd ..

 

학습 잘 된다.