λ°μΌνλ‘ λλ μμ°, λλ μλΉμ μλ. μ°λ¦¬λ λ§μ λ¬Όκ±΄μ΄ λλμΌλ‘ μμ°λκ³ , μλΉλλ μλλ₯Ό μ΄κ³ μμ΅λλ€. νμ§λ§ μ΄λ¬ν λ¬Ένλ 'μ°λ κΈ° λλ', 'λ§€λ¦½μ§ λΆμ‘±'κ³Ό κ°μ μ¬λ¬ μ¬ν λ¬Έμ λ₯Ό λ³κ³ μμ΅λλ€.
λΆλ¦¬μκ±°λ μ΄λ¬ν νκ²½ λΆλ΄μ μ€μΌ μ μλ λ°©λ² μ€ νλμ λλ€. μ λΆλ¦¬λ°°μΆ λ μ°λ κΈ°λ μμμΌλ‘μ κ°μΉλ₯Ό μΈμ λ°μ μ¬νμ©λμ§λ§, μλͺ» λΆλ¦¬λ°°μΆ λλ©΄ κ·Έλλ‘ νκΈ°λ¬Όλ‘ λΆλ₯λμ΄ λ§€λ¦½ λλ μκ°λκΈ° λλ¬Έμ λλ€.
λ°λΌμ μ°λ¦¬λ μ¬μ§μμ μ°λ κΈ°λ₯Ό Detection νλ λͺ¨λΈμ λ§λ€μ΄ μ΄λ¬ν λ¬Έμ μ μ ν΄κ²°ν΄λ³΄κ³ μ ν©λλ€. λ¬Έμ ν΄κ²°μ μν λ°μ΄ν°μ μΌλ‘λ μΌλ° μ°λ κΈ°, νλΌμ€ν±, μ’ μ΄, μ 리 λ± 10 μ’ λ₯μ μ°λ κΈ°κ° μ°ν μ¬μ§ λ°μ΄ν°μ μ΄ μ 곡λ©λλ€.
μ¬λ¬λΆμ μν΄ λ§λ€μ΄μ§ μ°μν μ±λ₯μ λͺ¨λΈμ μ°λ κΈ°μ₯μ μ€μΉλμ΄ μ νν λΆλ¦¬μκ±°λ₯Ό λκ±°λ, μ΄λ¦°μμ΄λ€μ λΆλ¦¬μκ±° κ΅μ‘ λ±μ μ¬μ©λ μ μμ κ²μ λλ€. λΆλ μ§κ΅¬λ₯Ό μκΈ°λ‘λΆν° ꡬν΄μ£ΌμΈμ! π
- μ 체 μ΄λ―Έμ§ κ°μ: 3272μ₯
- Train: 7617μ₯
- Validation: 655μ₯
- 11 classes:
Background, General trash, Paper, Paper pack, Metal, Glass, Plastic, Styrofoam, Plastic bag, Battery, Clothing
- Image size: (512x512)
- Coco formatμ annotation file
- Pixel μ’νμ λ°λΌ μΉ΄ν κ³ λ¦¬ κ°μ 리ν΄νμ¬ submission μμμ λ§κ² csv νμΌμ λ§λ€μ΄ μ μΆ
- Test setμ mIoU (Mean Intersection over Union)λ‘ νκ°
- Semantic Segmentationμμ μ¬μ©λλ λνμ μΈ μ±λ₯ μΈ‘μ λ°©λ²
- IoU
$IoU={|X \cap Y| \over |X \cup Y|}={|X \cap Y| \over {|X|+|Y|-|X \cap Y|}}$ - mIoU where
$c=11$ $mIoU={1 \over c} \sum_{c=1}^c{|X_c \cap Y_c|\over|X_c \cup Y_c|}$
κΆνμ°
Β μ΅μ’ νλ‘μ νΈ κΈ°νμ μμ± / νμ΅ λ°μ΄ν°μ μμ± / λͺ¨λΈ νμ΅κΉλμ
Β mmSegmentation νκ²½ μ€μ / λ°μ΄ν° μ μ 리 / λͺ¨λΈ νμ΅κΉμ°¬λ―Ό
Β baseline μ½λ μ€ν λ° λΆμμ΄μμ§
Β λ°μ΄ν°μ ν¬λ§·μ λ§κ² λ³κ²½ / mmsegmentationμ ν΅ν λͺ¨λΈ νμ΅ / pseudo labellingμ ν¨μ¬
Β baseline μ½λ μ€ν λ° λΆμ / λͺ¨λΈ νμ΅
.
βββ mmsegmentation
β βββ _custom_configs_
β βββ deeplabv3_r101_d8
β βββ hr_city_scape
β βββ ocr_hr18
β βββ upernet_beit_large_Albu
βββ utils
βββ stratified_kfold
β βββ copy_images_kfold.py
β βββ create_kfold.py
β βββ create_mask_kfold.py
βββ copy_images.ipynb
βββ create_mask.ipynb
βββ inference.py
βββ mask_mmseg_dataset.ipynb
βββ pseudo_labelling.ipynb
βββ train_val_to_coco.ipynb
/opt/ml/input
βββ data
β βββ batch_01_vt
β βββ batch_02_vt
β βββ batch_03
β βββ train.json
β βββ train_all.json
β βββ val.json
β βββ test.json
βββ mmseg
βββ annotations
β βββ train
β βββ val
βββ images
β βββ train
β βββ val
βββ test
cd mmsegmentation
python tools/train.py _custom_configs_/{μ¬μ©ν λͺ¨λΈ}/model.py
Decoder | Backbone | Parameter | LB mIoU |
---|---|---|---|
FCN | ResNet-50 | baseline | 0.5141 |
UNet++ | Conv block | baseline | 0.2181 |
HRNet-18 | OCRNet-18 | epoch 40, albumentations, TTA | 0.6148 |
Deeplabv3 | ResNet-101 | epoch 50, albumentations, TTA | 0.6429 |
UperNet | BEiT-L | epoch 50, albumentations, multi-scale, TTA | 0.8008 |
UperNet | Swin-L | epoch 50, albumentations, TTA | 0.7054 |
UperNet | BEiT-L | epoch 50, albumentations, TTA | 0.7773 |
UperNet | BEiT-L | epoch 50, albumentations, TTA, CLAHE | 0.7671 |
Senformer | Swin-L | epoch 50, albumentations, TTA | 0.6829 |
Upernet BEiT-L
+ Upernet Swin-L
= 0.8044
β Final Score
- Public LB score: mIoU 0.8044 (7λ±)
- Private LB score: mIoU 0.7551 (5λ±)