pretrained model 링크 쭉: https://github.com/microsoft/Swin-Transformer?tab=readme-ov-file

모델 하이퍼파라미터: https://github.com/microsoft/Swin-Transformer/blob/main/configs/swin/swin_tiny_patch4_window7_224_22k.yaml

실행파일

# 갱장히 간단하주
from mmengine.config import Config
from mmdet.utils import register_all_modules
from mmengine.runner import Runner

register_all_modules()

cfg = Config.fromfile('/data/ephemeral/home/hobbang/level2-objectdetection-cv-16/work_dirs/configs/hihi.py')

cfg.work_dir = '/data/ephemeral/home/hobbang/level2-objectdetection-cv-16/work_dirs/swin'

runner = Runner.from_cfg(cfg)
runner.train()

hihi.py

# 상속 방법이 굉장히 신기하주
_base_ = '/data/ephemeral/home/hobbang/level2-objectdetection-cv-16/baseline/mmdetection/configs/swin/retinanet_swin-t-p4-w7_fpn_1x_coco.py'

model = dict(
    backbone=dict(
        norm_cfg=dict(type='LN', requires_grad=False)
    ),
    bbox_head=dict(
        num_classes=10
    )
)

train_dataloader = dict(
    batch_size=2
)

optim_wrapper = dict(optimizer=dict(lr=0.001))

# 초반에 크게 lr이 오르는건 linearLR로 warmup해서 그런거

coco_detection.py: 데이터셋은 원본을 수정하는게 편할듯?

# dataset settings
dataset_type = 'CocoDataset'
data_root = '/data/ephemeral/home/hobbang/level2-objectdetection-cv-16/dataset/'
classes = ("General trash", "Paper", "Paper pack", "Metal", "Glass", "Plastic", "Styrofoam", "Plastic bag", "Battery", "Clothing")

# Example to use different file client
# Method 1: simply set the data root and let the file I/O module
# automatically infer from prefix (not support LMDB and Memcache yet)

# data_root = 's3://openmmlab/datasets/detection/coco/'

# Method 2: Use `backend_args`, `file_client_args` in versions before 3.0.0rc6
# backend_args = dict(
#     backend='petrel',
#     path_mapping=dict({
#         './data/': 's3://openmmlab/datasets/detection/',
#         'data/': 's3://openmmlab/datasets/detection/'
#     })
backend_args = None

train_pipeline = [
    dict(type='LoadImageFromFile', backend_args=backend_args),
    dict(type='LoadAnnotations', with_bbox=True),
    dict(type='Resize', scale=(1333, 800), keep_ratio=True),
    dict(type='RandomFlip', prob=0.5),
    dict(type='PackDetInputs')
]
test_pipeline = [
    dict(type='LoadImageFromFile', backend_args=backend_args),
    dict(type='Resize', scale=(1333, 800), keep_ratio=True),
    # If you don't have a gt annotation, delete the pipeline
    dict(type='LoadAnnotations', with_bbox=True),
    dict(
        type='PackDetInputs',
        meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
                   'scale_factor'))
]
train_dataloader = dict(
    batch_size=2,
    num_workers=2,
    persistent_workers=True,
    sampler=dict(type='DefaultSampler', shuffle=True),
    batch_sampler=dict(type='AspectRatioBatchSampler'),
    dataset=dict(
        type=dataset_type,
        data_root=data_root,
        ann_file='train.json',
        data_prefix=dict(img=''),
        filter_cfg=dict(filter_empty_gt=True, min_size=32),
        pipeline=train_pipeline,
        backend_args=backend_args,
        metainfo=dict(classes=classes)
        ))
val_dataloader = dict(
    batch_size=1,
    num_workers=2,
    persistent_workers=True,
    drop_last=False,
    sampler=dict(type='DefaultSampler', shuffle=False),
    dataset=dict(
        type=dataset_type,
        data_root=data_root,
        ann_file='train.json',
        data_prefix=dict(img=''),
        metainfo=dict(classes=classes),
        test_mode=True,
        pipeline=test_pipeline,
        backend_args=backend_args))
test_dataloader = val_dataloader

val_evaluator = dict(
    type='CocoMetric',
    ann_file=data_root + 'train.json',
    metric='bbox',
    format_only=False,
    backend_args=backend_args,
    # classwise=True # classwise: 평가 지표를 개별적으로 계산할지 선택. True면 클래스별 분석 가능, False면 전체적인 성능 파악에 도움이 됨
    )
# test_evaluator = ~~
test_evaluator = val_evaluator

# inference on test dataset and
# format the output results for submission.
# test_dataloader = dict(
#     batch_size=1,
#     num_workers=2,
#     persistent_workers=True,
#     drop_last=False,
#     sampler=dict(type='DefaultSampler', shuffle=False),
#     dataset=dict(
#         type=dataset_type,
#         data_root=data_root,
#         ann_file=data_root + 'annotations/image_info_test-dev2017.json',
#         data_prefix=dict(img='test2017/'),
#         test_mode=True,
#         pipeline=test_pipeline))
# test_evaluator = dict(
#     type='CocoMetric',
#     metric='bbox',
#     format_only=True,
#     ann_file=data_root + 'annotations/image_info_test-dev2017.json',
#     outfile_prefix='./work_dirs/coco_detection/test')