์ ๋ค์ด | ๋ฐ์ ์ฌ | ๊น๊ท๋ฏผ | ์ด์ตํฌ | ์๊ฒฝ๊ตญ |
---|---|---|---|---|
์ ๋ค์ด
ConvNeXt โข BEiT โข Model Experiment โข Git management ๋ฐ์ ์ฌ
Data Preprocessing โข SwinL โข Model Experiment โข Pseudo Labeling โข Stratified Kfold โข Ensemble ๊น๊ท๋ฏผ
Model Experiment โข Loss โข Hyper Parameter Tuning์ด์ตํฌ
Model Experiment โข BEiT โข SwinL โข Ensemble ์๊ฒฝ๊ตญ
Model Experiment โข UNet++ โข Segformerm โข HRNetV2 โข SwinL โข Pseudo Labeling[์ฐธ๊ณ ์ฌํญ]
model๋ก๋ถํฐ ์์ธก๋ mask์ size๋ 512 x 512 ์ง๋ง, ๋ํ์ ์ํํ ์ด์์ ์ํด output์ ์ผ๊ด์ ์ผ๋ก 256 x 256 ์ผ๋ก ๋ณ๊ฒฝํ์ฌ score๋ฅผ ๋ฐ์ํ๊ฒ ๋์์ต๋๋ค.
mIoU (Mean Intersection over Union)
ํ์ดํ์ดํ_๋ฐํ์๋ฃ.pdf
.
|-- code (git)
| |-- examples
| | |-- baseline_model_feat
| | | |-- bestmIoU_baseline_fcn_resnet50.ipynb
| | | `-- wandb_baseline_fcn_resnet50.ipynb
| | `-- data
| | |-- convert_mmseg.ipynb
| | |-- data_concat_anno_exclude.ipynb
| | |-- data_concat_img_exclude.ipynb
| | `-- stratified_kfold.py
| |-- mmsegmentation
| |-- model
| | `-- mmseg
| |-- baseline_fcn_resnet50.ipynb
| |-- requirements.txt
| `-- utils.py
`-- data
|-- batch_01_vt
| `-- 0002.jpg
|-- batch_02_vt
| `-- 0001.jpg
|-- batch_03
| `-- 0001.jpg
|-- mmseg
| |-- ann_dir
| | |-- train0~4
| | | `-- *.png
| | `-- val0~4
| | `-- *.png
| `-- img_dir
| |-- test
| | `-- *.jpg
| |-- train0~4
| | `-- *.jpg
| `-- val0~4
| `-- *.jpg
|-- stratified_kfold
| |-- train0~4.json
| `-- val0~4.json
|-- new_train_all_anno_excluded.json
|-- test.json
|-- train.json
|-- train_all.json
`-- val.json
์ค์น ์์
conda create -n open-mmlab python=3.10 -y
conda activate open-mmlab
conda install pytorch=1.11.0 torchvision cudatoolkit=11.3 -c pytorch
pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu113/torch1.11/index.html
# mmsegmentation์ ์ค์นํ ๊ฒฝ๋ก ์ค์ . ์ ๋ ์ ํฌ ๊น ํด๋ ๋ด๋ถ์ ์ค์นํ์ต๋๋ค.
cd ... code/
git clone https://github.com/open-mmlab/mmsegmentation.git
cd mmsegmentation
pip install -e . # or "python setup.py develop"