κΉλ²μ€ | λ°±μ°μ΄ | μ‘°μ©μ¬ | μ‘°μ€μ¬ | μ΅λͺ ν |
---|---|---|---|---|
λΆλ¦¬μκ±°λ μ΄λ¬ν νκ²½ λΆλ΄μ μ€μΌ μ μλ λ°©λ² μ€ νλλ€. 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΄λ€. λ°λΌμ μ¬μ§μμ μ°λ κΈ°λ₯Ό Segmentationνλ λͺ¨λΈμ λ§λ€μ΄ μ΄λ¬ν λ¬Έμ μ μ ν΄κ²°ν΄λ³΄κ³ μ νλ€. λ¬Έμ ν΄κ²°μ μν λ°μ΄ν°μ μΌλ‘ 11κ°μ§μ μ°λ κΈ°κ° μ°ν μ¬μ§ λ°μ΄ν°μ μ΄ μ 곡λλ€.
Dataset
Background
,Β Generaltrash
,Β Paper
,Β Paperpack
,Β Metal
,Β Glass
,Β Plastic
,Β Styrofoam
,Β Plastic bag
,Β Battery
,Β Clothing
id
,height
,width
,filename
id
,segmentation
,bbox
,area
,category_id
,image_id
[2022.12.19 - 2022.12.21]
[2022.12.21 - 2022.12.31]
SweepμΌλ‘ λͺ¨λΈ 체리νΌνΉ (smp μ¬μ©)
[2023.01.01 - 2023.01.05]
[2023.01.02 - 2023.01.05]
mIoU | Ranking | |
---|---|---|
PUBLIC | 0.7665 | 9th |
PRIVATE | 0.7563 | 8th |
Step 1: Train
ex)
python mmsegmentation/tools/train.py \ _trashsegmentation/convnext/upernet_convnext_xlarge_fp16_640x640_160k_ade20k.py
# configμλ μνλ νμΌ κ²½λ‘λ₯Ό λ£μ΄μ€λ€.
python mmsegmentation/_trashsegmentation/utils/inference.py
python smp/train.py
python smp/inference.py
cd ViT-Adapter/segmentation
Step 1: Train
ex)
python train.py configs/upstage/mask2former_beit_adapter_base_512_upstage_ss.py
# configμλ μνλ νμΌ κ²½λ‘λ₯Ό λ£μ΄μ€λ€.
inference.ipynb μ΄μ©
π level2_semanticsegmentation_cv-level2-cv-01
βββ π copy_paste
β βββ π src
β β βββ π create_annotations.py
β βββ π concatjson.ipynb
β βββ π copy_paste.py
β βββ π create-custom-coco-dataset.ipynb
β βββ π get_coco_mask.ipynb
β
βββ π images
β βββ π Augmentation_img-1.py
β βββ π Augmentation_img-2.py
β βββ π dataset.png
β βββ π description.png
β βββ π Timeline.png
β
βββ π mmsegmentation
β βββ π _trashsegmentation
β β βββ π __base__
β β β βββ π datasets
β β β β βββ π upstage.py
β β β β
β β β βββ π models
β β β β βββ π segformer_mit-b0.py
β β β β βββ π segmenter_vit-b16_mask.py
β β β β βββ π upernet_beit.py
β β β β βββ π upernet_convnext.py
β β β β βββ π upernet_swin.py
β β β β
β β β βββ π schedules
β β β β βββ π schedule_160k.py
β β β β βββ π schedule.py
β β β β
β β β βββ π default_runtime.py
β β β
β β βββ π beit
β β β βββ π upernet_beit-base_8x2_640x640_160k_ade20k.py
β β β βββ π upernet_beit-base_640x640_160k_ade20k_ms.py
β β β βββ π upernet_beit-large_fp16_8x1_640x640_160k_ade20k.py
β β β βββ π upernet_beit-large_fp16_640x640_160k_ade20k_ms.py
β β β
β β βββ π convnext
β β β βββ π upernet_convnext_base_fp16_512x512_160k_ade20k.py
β β β βββ π upernet_convnext_xlarge_fp16_640x640_160k_ade20k.py
β β β
β β βββ π segformer
β β β βββ π segformer_mit-b0_512x512_160k_ade20k.py
β β β βββ π segformer_mit-b5_512x512_160k_ade20k.py
β β β
β β βββ π segmenter
β β β βββ π segmenter_vit-l_mask_8x1_640x640_160k_ade20k.py
β β β
β β βββ π swin
β β β βββ π upernet_swin_large_patch4_window7_512x512_pretrain_224x224_22K_160k_ade20k.py
β β β βββ π upernet_swin_large_patch4_window12_512x512_pretrain_384x384_22K_160k_ade20k.py
β β β βββ π upernet_swin_tiny_patch4_window7_512x512_160k_ade20k_pretrain_224x224_1K.py
β β β
β β βββ π utils
β β βββ π convert2mmseg.py
β β βββ π inference.ipynb
β β βββ π inference.py
β β
β βββ π config
β β
β βββ π mmseg
β β
β βββ π requirements
β β
β βββ π submission
β β
β βββ π tests
β β
β βββ π tools
β βββ π train.py
β βββ π test.py
β βββ π convert_datasets
β βββ π model_converters
β βββ ...
β
βββ π smp
β βββ π submission
β βββ π dataset.py
β βββ π eval.py
β βββ π inference.py
β βββ π requirements.txt
β βββ π train.py
β βββ π utils.py
β
β
βββ π ViT-Adapter
β βββ π segmentation
β β βββ π configs
β β β βββ π _base_
β β β β βββ π datasets
β β β β β βββ π upstage.py
β β β β β
β β β β βββ π models
β β β β β βββ π mask2former_beit_upstage.py
β β β β β βββ π upernet_beit_upstage.py
β β β β β
β β β β βββ π schedules
β β β β β βββ π schedule.py
β β β β β βββ π schedule_fp16.py
β β β β β
β β β β βββ π default_runtime.py
β β β β
β β β βββ π upstage
β β β βββ π mask2former_beit_adapter_base_512_upstage_ss.py
β β β βββ π mask2former_beit_adapter_large_512_upstage_ss.py
β β β βββ π upernet_augreg_adapter_base_512_160k_upstage.py
β β β βββ π upernet_deit_adapter_tiny_512_160k_upstage.py
β β β
β β βββ π mmcv_custom
β β βββ π mmseg_custom
β β βββ π inference.ipynb
β β βββ π train.py
β β βββ π train_fp16.py
β β βββ ...
β βββ π ...
β
β
βββ π utils
β βββ π 5fold.ipynb
β βββ π cleansing_by_feat.py
β βββ π cleansing_by_name.py
β βββ π combine_coco.py
β βββ π eda.py
β βββ π post_processing.py
β βββ π size.png
β
βββ π README.md