How can we not evaluate the background class in voc2012 dataset

[qiuzhixin9527]

Don't care about background class

[TianWen580]

The easiest way is to just ignore it but about the class 'background' is a lot to say(ref from OpenMMLab):
When your ground-truth contains the class 'background', you can set the reduce_zero_label to True in the train pipeline, so that the background have no way to be trained and the evaluation result of background will disappeared then. Here is an example(notice the param reduce-zero-label=True)👇

# xxxconfigs.py

...
train_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(type='LoadAnnotations', reduce_zero_label=True),
    dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),
    dict(type='RandomCrop', crop_size=crop_size, cat_max_ratio=0.75),
    dict(type='RandomFlip', prob=0.5),
    dict(type='PhotoMetricDistortion'),
    dict(type='Normalize', **img_norm_cfg),
    dict(type='Pad', size=crop_size, pad_val=0, seg_pad_val=255),
    dict(type='DefaultFormatBundle'),
    dict(type='Collect', keys=['img', 'gt_semantic_seg']),
]
...

Meanwhile, you are required to set CLASSES to ('stamp', ) and to set num_classes to 1 so that everything is suitable.

---

最简单的方法是忽略它，但关于“背景”类别有很多可以说的(参考文章)：
当你的ground-truth包含“背景”类时，你可以在train pipeline中将reduce_zero_label设置为True，这样背景就不会参与训练，背景的评估结果也将消失。举个配置文件的栗子（注意param reduce-zero-label=True）👇

# xxxconfigs.py

...
train_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(type='LoadAnnotations', reduce_zero_label=True),
    dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),
    dict(type='RandomCrop', crop_size=crop_size, cat_max_ratio=0.75),
    dict(type='RandomFlip', prob=0.5),
    dict(type='PhotoMetricDistortion'),
    dict(type='Normalize', **img_norm_cfg),
    dict(type='Pad', size=crop_size, pad_val=0, seg_pad_val=255),
    dict(type='DefaultFormatBundle'),
    dict(type='Collect', keys=['img', 'gt_semantic_seg']),
]
...

同时，你需要将 CLASSES 设置为 ('stamp', )，并将 num_classes 设置为1，使参数都对得上。