coco_detection.py 代码注释

coco_detection.py 代码注释

# dataset settings
dataset_type = 'CocoDataset'
#数据集路径
data_root = 'F:\\data\\coco/'#data/coco/
#归一化参数
img_norm_cfg = dict(
    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
#train流水线
train_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(type='LoadAnnotations', with_bbox=True),
    #1333*800：如果有一张100*100的图，经过这一行后会变成800*800的
    #因为要保证宽高比，开始1:1，转化后还是1:1
    #如果是1000*1000则变为1333*1333（1280*1280）
    #要是32的倍数
    dict(type='Resize', img_scale=(1333, 800), keep_ratio=True),
    dict(type='RandomFlip', flip_ratio=0.5),
    dict(type='Normalize', **img_norm_cfg),
    dict(type='Pad', size_divisor=32),
    dict(type='DefaultFormatBundle'),
    dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels']),#收集三类信息
]
#test流水线
test_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(
        type='MultiScaleFlipAug',
        img_scale=(1333, 800),
        flip=False,
        transforms=[
            dict(type='Resize', keep_ratio=True),
            dict(type='RandomFlip'),
            dict(type='Normalize', **img_norm_cfg),
            dict(type='Pad', size_divisor=32),
            dict(type='ImageToTensor', keys=['img']),
            dict(type='Collect', keys=['img']),
        ])
]
#数据
data = dict(
    samples_per_gpu=2,#每个GPU放多少samples，相当于batch_size，根据现存大小进行调整
    workers_per_gpu=2,#每个GPU放多少workers
    train=dict(
        type=dataset_type,
        ann_file=data_root + 'annotations/instances_train2017.json',#标签路径
        img_prefix=data_root + 'train2017/',#图片路径
        pipeline=train_pipeline),
    val=dict(
        type=dataset_type,
        ann_file=data_root + 'annotations/instances_val2017.json',#标签路径
        img_prefix=data_root + 'val2017/',#图片路径
        pipeline=test_pipeline),
    test=dict(
        type=dataset_type,
        ann_file=data_root + 'annotations/instances_val2017.json',#标签路径
        img_prefix=data_root + 'val2017/',#图片路径
        pipeline=test_pipeline))
#interval间隔：1，每过1个epoch进行验证
evaluation = dict(interval=1, metric='bbox')