λλ μμ°, λλ μλΉμ μλμμ βμ°λ κΈ° λλβ, βλ§€λ¦½μ§ λΆμ‘±βκ³Ό κ°μ μ¬ν λ¬Έμ κ° λ°μνκ³ μλ€. μ΄λ¬ν λ¬Έμ λ₯Ό ν΄κ²°νκΈ° μν΄μ λΆλ¦¬μκ±°λ νκ²½ λΆλ΄μ μ€μΌ μ μλ λ°©λ²μ΄λ€. μ λΆλ¦¬μκ±° λ μ°λ κΈ°λ μμμΌλ‘μ κ°μΉλ₯Ό μΈμ λ°μ μ¬νμ©λμ§λ§, μλͺ» λΆλ¦¬λ°°μΆ λλ©΄ νκΈ°λ¬Όλ‘ λΆλ₯λμ΄ λ§€λ¦½ λλ μκ°λλ€.
λ°λΌμ μ°λ κΈ° μ¬μ§μ ν΅ν΄ 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©ν μ μλ μμ€ν μ΄ νμνλ€. λ³Έ νλ‘μ νΈμμλ μΌλ° μ°λ κΈ°, νλΌμ€ν±, μ’ μ΄, μ 리 λ± 10μ’ λ₯μ μ°λ κΈ°λ₯Ό λΆλ₯νλ Object Detectionμ μννλ€.
-
λ°μ΄ν°μ κ°μ : Train 4,883μ₯ / Test 4,871μ₯
-
μ΄λ―Έμ§ ν¬κΈ° : (1024, 1024)
-
ν΄λμ€μ λΆλ₯
- General trash
- Paper
- Paper pack
- Metal
- Glass
- Plastic
- Styrofoam
- Plastic bag
- Battery
- Clothing
νμ΅ν μμ 7κ°μ λͺ¨λΈμ inference κ²°κ³Όλ₯Ό WBF(Weighted Boxes Function)μ μ¬μ©νμ¬ μ μΆν inference κ²°κ³Όκ° κ°μ₯ λμ scoreλ₯Ό λ¬μ±.
-
μμλΈμ μ¬μ©ν λͺ¨λΈ
idx model mAP 50 1 SwinTransformer Large - FasterRCNN 0.5893 2 SwinTransformer Large - FasterRCNN 0.5949 3 SwinTransformer Large - FasterRCNN 0.6072 4 SwinTransformer Large - FasterRCNN 0.6100 5 SwinTransformer Large - FasterRCNN 0.6176 6 Yolov7 0.5633 7 EfficientDet 0.5556 -
μ΅μ’ κ²°κ³Ό
mAP : 0.6753
κΉλμΈ_T4029 | μ μμ€_T4113 | μ΄νμ§_T4177 | μ νκΈ°_T4205 | νμ€ν_T4235 |
Model experiments, Ensemble | EDA, Data Augmentation | Model experiments | EDA, Data Augmentation | Model experiments, Data Augmentation |