测试集处理

[huisuli]

在train.py中为什么加载数据这里val_pipeline = copy.deepcopy(train_pipeline) 测试集的处理需要复制训练集的处理，这样训练集每次都会随机裁剪尺寸，在evaluation.py里也是这样，
train_dataset = Mydataset(train_datas, train_pipeline)
val_pipeline = copy.deepcopy(train_pipeline)
val_dataset = Mydataset(val_datas, val_pipeline)
train_loader = DataLoader(train_dataset, shuffle=True, batch_size=data_cfg.get('batch_size'), num_workers=data_cfg.get('num_workers'),pin_memory=True, drop_last=True, collate_fn=collate)
val_loader = DataLoader(val_dataset, shuffle=False, batch_size=data_cfg.get('batch_size'), num_workers=data_cfg.get('num_workers'), pin_memory=True,
drop_last=True, collate_fn=collate)

[Fafa-DL]

是的，当时官方也是这样做的。如果测试时固定大小我也有做过，精度会有一定下降。建议多次取均值，下一次更新考虑自主选择测试方法

[huisuli]

那我也可以把测试集这部分加上标签，送进去训练和测试，这样也行吧？
val_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='CenterCrop',
crop_size=384,
efficientnet_style=True,
interpolation='bicubic'),
dict(type='Normalize', **img_norm_cfg),
dict(type='ImageToTensor', keys=['img']),
dict(type='ToTensor', keys=['gt_label']),
dict(type='Collect', keys=['img', 'gt_label'])
]

[MadMrFox]

测试时将训练参数设置不做任何增强和裁剪就可以稳定输出了，唯一的问题是，训练时进行了增强数据，测试时不做处理的结果会差很多。
以resnet50.py为例
原本的
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='RandAugment',
policies=policies,
num_policies=2,
magnitude_level=12),
dict(
type='RandomResizedCrop',
size=224,
efficientnet_style=True,
interpolation='bicubic',
backend='pillow'),
dict(type='RandomFlip', flip_prob=0.5, direction='horizontal'),
dict(type='ColorJitter', brightness=0.4, contrast=0.4, saturation=0.4),
dict(type='Lighting', **img_lighting_cfg),
dict(
type='Normalize',
mean=[123.675, 116.28, 103.53],
std=[58.395, 57.12, 57.375],
to_rgb=False),
dict(type='ImageToTensor', keys=['img']),
dict(type='ToTensor', keys=['gt_label']),
dict(type='Collect', keys=['img', 'gt_label'])
]
改为
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='Resize', size=224),
dict(
type='Normalize',
mean=[123.675, 116.28, 103.53],
std=[58.395, 57.12, 57.375],
to_rgb=False),
dict(type='ImageToTensor', keys=['img']),
dict(type='ToTensor', keys=['gt_label']),
dict(type='Collect', keys=['img', 'gt_label'])
]

[Fafa-DL]

已修复评估时结果不稳定问题，可拉取最新代码