Fashion semantic segmentation(feat. SCHP 성공)

2 min readJun 12, 2021

conda env create -f environment.yaml여기서 근데 pytorch 1.5.1 버전을 설치하네.. 내꺼 RTX3090이라서 호환이안될텐데;;conda env create -f environment.yaml
conda activate schp
pip install -r requirements.txt
mkdir checkpoints
mkdir inputs
mkdir outputs
pip install gdown
python 키고
import gdown
if dataset == ‘lip’:
 url = ‘https://drive.google.com/uc?id=1k4dllHpu0bdx38J7H28rVVLpU-kOHmnH'
elif dataset == ‘atr’:
 url = ‘https://drive.google.com/uc?id=1ruJg4lqR_jgQPj-9K0PP-L2vJERYOxLP'
elif dataset == ‘pascal’:
 url = ‘https://drive.google.com/uc?id=1E5YwNKW2VOEayK9mWCS3Kpsxf-3z04ZE'
output = ‘checkpoints/final.pth’
gdown.download(url, output, quiet=False)

그럼 checkpoints 파일에 final.pth가 있는걸 볼 수 있다.

Run: python simple_extractor.py — dataset ‘lip’ — model-restore ‘checkpoints/final.pth’ — input-dir inputs/ — output-dir outputs/

Error: OSError: CUDA_HOME environment variable is not set. Please set it to your CUDA install root.

쿠다 패스 설정해야함.

environment.yaml 여기에 쿠다설치되어있는데보니깐 prefix: /home/peike/opt/anaconda3/envs/schp 이렇게 되어있네.prefix: /data/juhwan/anaconda3/envs/schp/ 로 바꾸고 다시돌림.

그냥 다삭제하고 다시 설치

conda remove -n schp — all
rm -rf Self-Correction-Human-Parsing/

아 이게아닌가..? 에러똑같네. CUDA home_path설정하자.

Run: CUDA_HOME=/data/juhwan/anaconda3/envs/schp/lib/python3.8/site-packages/torch/cuda python3 simple_extractor.py — dataset ‘lip’ — model-restore ‘checkpoints/final.pth’ — input-dir ‘inputs’ — output-dir ‘outputs’

Error:ValueError: Unknown CUDA arch (8.6) or GPU not supported

역시나 맨처음 pytorch, cuda 를 버전을 GTX3090이랑 안되서 , 잘못잡아서그렇다.

conda uninstall pytorch
conda uninstall libtorch

conda install pytorch torchvision torchaudio cudatoolkit=11.1 -c pytorch -c nvidia

.
CUDA_HOME=/data/juhwan/anaconda3/envs/schp/lib/python3.8/site-packages/torch/cuda python3 simple_extractor.py --dataset 'lip' --model-restore 'checkpoints/final.pth' --input-dir 'inputs' --output-dir 'outputs'

ninja: build stopped: subcommand failed.

하.. 그냥 colab 사용할래

Colab