MMSegmentation系列之训练与推理自己的数据集（三）

1、准备数据目录结构

mstsc
1

VOCdevkit
│ │ ├── VOC2012
│ │ │ ├── JPEGImages（原始影像）
│ │ │ ├── SegmentationClass（掩膜影像）
│ │ │ ├── ImageSets
│ │ │ │ ├── Segmentation（数据划分）

2、下载预训练模型

deeplabv3plus

3、修改配置文件（deeplabv3plus_r50-d8_512x512_40k_voc12aug.py）

1、 设置修改类别数​(模型架构配置文件deeplabv3plus_r50-d8_512x512_40k_voc12aug.py)

_base_ = [
    '../_base_/models/deeplabv3plus_r50-d8.py',
    '../_base_/datasets/pascal_voc12_aug.py', '../_base_/default_runtime.py',
    '../_base_/schedules/schedule_40k.py'
]
model = dict(
    decode_head=dict(num_classes=21), auxiliary_head=dict(num_classes=21))#按照自己的数据进行修改
1
2
3
4
5
6
7

2、修改数据信息（数据类型、数据主路径等和batch-size）（…/base/datasets/pascal_voc12_aug.py、pascal_voc12.py）

# dataset settings
dataset_type = 'PascalVOCDataset' #按照需要修改数据类型
data_root = 'data/VOCdevkit/VOC2012' #按照需要修改数据主路径
img_norm_cfg = dict(
    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
crop_size = (512, 512)
train_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(type='LoadAnnotations'),
    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']),
]
test_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(
        type='MultiScaleFlipAug',
        img_scale=(2048, 512),
        # img_ratios=[0.5, 0.75, 1.0, 1.25, 1.5, 1.75],
        flip=False,
        transforms=[
            dict(type='Resize', keep_ratio=True),
            dict(type='RandomFlip'),
            dict(type='Normalize', **img_norm_cfg),
            dict(type='ImageToTensor', keys=['img']),
            dict(type='Collect', keys=['img']),
        ])
]
data = dict(
    samples_per_gpu=4, #按照硬件设备修改
    workers_per_gpu=4,
    train=dict(
        type=dataset_type,
        data_root=data_root,
        img_dir='JPEGImages', #同理按照需要进行数据修改
        ann_dir='SegmentationClass',
        split='ImageSets/Segmentation/train.txt',
        pipeline=train_pipeline),
    val=dict(
        type=dataset_type,
        data_root=data_root,
        img_dir='JPEGImages',
        ann_dir='SegmentationClass',
        split='ImageSets/Segmentation/val.txt',
        pipeline=test_pipeline),
    test=dict(
        type=dataset_type,
        data_root=data_root,
        img_dir='JPEGImages',
        ann_dir='SegmentationClass',
        split='ImageSets/Segmentation/val.txt',
        pipeline=test_pipeline))
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57

_base_ = './pascal_voc12.py'
# dataset settings,因为我们没有Aug数据所以频闭掉
#data = dict(
   # train=dict(
       # ann_dir=['SegmentationClass', 'SegmentationClassAug'],
        #split=[
           # 'ImageSets/Segmentation/train.txt',
           # 'ImageSets/Segmentation/aug.txt'
       # ]))

1
2
3
4
5
6
7
8
9
10

重要:配置文件中的默认学习速率为4 gpu和2 img/gpu(批处理大小= 4x2 = 8)。同样地，你也可以使用8 gpu和1 imgs/gpu，因为所有型号都使用cross_gpu SyncBN。

3、修该类别名称CLASSES（mmseg/datasets/voc.py）

class PascalVOCDataset(CustomDataset):
    """Pascal VOC dataset.

    Args:
        split (str): Split txt file for Pascal VOC.
    """

    CLASSES = ('background', 'aeroplane', 'bicycle', 'bird', 'boat', 'bottle',
               'bus', 'car', 'cat', 'chair', 'cow', 'diningtable', 'dog',
               'horse', 'motorbike', 'person', 'pottedplant', 'sheep', 'sofa',
               'train', 'tvmonitor') #修改为自己的数据类别
1
2
3
4
5
6
7
8
9
10
11

4、修改运行信息配置（加载预训练模型和断点训练）（configs/-base-/default_runtime.py）

log_config = dict(
    interval=50,
    hooks=[
        dict(type='TextLoggerHook', by_epoch=False),
        # dict(type='TensorboardLoggerHook')
    ])
# yapf:enable
dist_params = dict(backend='nccl')
log_level = 'INFO'
load_from = None  #可以加载预训练模型
resume_from = None  #可以加载断点训练模型
workflow = [('train', 1)]
cudnn_benchmark = True
1
2
3
4
5
6
7
8
9
10
11
12
13

5、修改运行信息配置（模型训练的最大次数、训练每个几次保留一个checkpoints、间隔多少次进行模型训练，模型训练评估的指标为、保留最好的模型）（configs/-base-/schedule_40k.py）

# optimizer
optimizer = dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0005)
optimizer_config = dict()
# learning policy
lr_config = dict(policy='poly', power=0.9, min_lr=1e-4, by_epoch=False)
# runtime settings
runner = dict(type='IterBasedRunner', max_iters=40000)#
checkpoint_config = dict(by_epoch=False, interval=4000)
evaluation = dict(interval=4000, metric='mIoU')#save_best='auto'

1
2
3
4
5
6
7
8
9
10

6、模型信息修改(tiicks)（configs/base/models）

Different Learning Rate(LR) for Backbone and Heads
n MMSegmentation, you may add following lines to config to make the LR of heads 10 times of backbone.

optimizer=dict(
    paramwise_cfg = dict(
        custom_keys={
            'head': dict(lr_mult=10.)}))
1
2
3
4
5
6
7

Online Hard Example Mining (OHEM)
We implement pixel sampler here for training sampling. Here is an example config of training PSPNet with OHEM enabled.

_base_ = './pspnet_r50-d8_512x1024_40k_cityscapes.py'
model=dict(
    decode_head=dict(
        sampler=dict(type='OHEMPixelSampler', thresh=0.7, min_kept=100000)) )
1
2
3
4
5
6
7

Class Balanced Loss
For dataset that is not balanced in classes distribution, you may change the loss weight of each class. Here is an example for cityscapes dataset.

_base_ = './pspnet_r50-d8_512x1024_40k_cityscapes.py'
model=dict(
    decode_head=dict(
        loss_decode=dict(
            type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0,
            # DeepLab used this class weight for cityscapes
            class_weight=[0.8373, 0.9180, 0.8660, 1.0345, 1.0166, 0.9969, 0.9754,
                        1.0489, 0.8786, 1.0023, 0.9539, 0.9843, 1.1116, 0.9037,
                        1.0865, 1.0955, 1.0865, 1.1529, 1.0507])))
1
2
3
4
5
6
7
8
9
10
11
12

Multiple Losses
For loss calculation, we support multiple losses training concurrently. Here is an example config of training unet on DRIVE dataset, whose loss function is 1:3 weighted sum of CrossEntropyLoss and DiceLoss:

_base_ = './fcn_unet_s5-d16_64x64_40k_drive.py'
model = dict(
    decode_head=dict(loss_decode=[dict(type='CrossEntropyLoss', loss_name='loss_ce', loss_weight=1.0),
            dict(type='DiceLoss', loss_name='loss_dice', loss_weight=3.0)]),
    auxiliary_head=dict(loss_decode=[dict(type='CrossEntropyLoss', loss_name='loss_ce',loss_weight=1.0),
            dict(type='DiceLoss', loss_name='loss_dice', loss_weight=3.0)]),
    )
1
2
3
4
5
6
7
8
9
10

Ignore specified label index in loss calculation
In default setting, avg_non_ignore=False which means each pixel counts for loss calculation although some of them belong to ignore-index labels.

For loss calculation, we support ignore index of certain label by avg_non_ignore and ignore_index. In this way, the average loss would only be calculated in non-ignored labels which may achieve better performance, and here is the reference. Here is an example config of training unet on Cityscapes dataset: in loss calculation it would ignore label 0 which is background and loss average is only calculated on non-ignore labels:

_base_ = './fcn_unet_s5-d16_4x4_512x1024_160k_cityscapes.py'
model = dict(
    decode_head=dict(
        ignore_index=0,
        loss_decode=dict(
            type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0, avg_non_ignore=True),
    auxiliary_head=dict(
        ignore_index=0,
        loss_decode=dict(
            type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0, avg_non_ignore=True)),
    ))
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16

4、模型进行训练

1、Train with a single GPU

python tools/train.py ${CONFIG_FILE} [optional arguments]
如果您想在命令中指定工作目录，您可以添加一个参数--work-dir ${YOUR_WORK_DIR}
python tools/train.py  configs/deeplabv3plus/deeplabv3plus_r50-d8_512x512_20k_voc12aug.py  --work-dir work_dirs/runs/train/deeplabv3plus

1
2
3
4

2、Train with multiple GPUs

sh tools/dist_train.sh ${CONFIG_FILE} ${GPU_NUM} [optional arguments]
1
–no-validate (not suggested): By default, the codebase will perform evaluation at every k iterations during the training. To disable this behavior, use --no-validate.
–work-dir ${WORK_DIR}: Override the working directory specified in the config file.
–resume-from ${CHECKPOINT_FILE}: Resume from a previous checkpoint file (to continue the training process).
–load-from ${CHECKPOINT_FILE}: Load weights from a checkpoint file (to start finetuning for another task).
–deterministic: Switch on “deterministic” mode which slows down training but the results are reproducible.
1
2
3
4
5
6
7

5、模型训练日志的打印划图

#Plot training logs
tools/analyze_logs.py plots loss/mIoU curves given a training log file
1
2

python tools/analyze_logs.py xxx.log.json [--keys ${KEYS}] [--legend ${LEGEND}] [--backend ${BACKEND}] [--style ${STYLE}] [--out ${OUT_FILE}]
1

Plot the mIoU, mAcc, aAcc metrics.
python tools/analyze_logs.py log.json --keys mIoU mAcc aAcc --legend mIoU mAcc aAcc
Plot loss metric.
python tools/analyze_logs.py log.json --keys loss --legend loss
1
2
3
4

6、模型测试

single-gpu testing

#单个模型测试
python tools/test.py ${CONFIG_FILE} ${CHECKPOINT_FILE} [--out ${RESULT_FILE}] [--eval ${EVAL_METRICS}] [--show]

#多个模型测试
/tools/dist_test.sh ${CONFIG_FILE} ${CHECKPOINT_FILE} ${GPU_NUM} [--out ${RESULT_FILE}] [--eval ${EVAL_METRICS}]
1
2
3
4
5

./tools/dist_test.sh configs/pspnet/pspnet_r50-d8_512x1024_40k_cityscapes.py   checkpoints/pspnet_r50-d8_512x1024_40k_cityscapes_20200605_003338-2966598c.pth   4 --out results.pkl --eval mIoU cityscapes


./tools/dist_test.sh \
    configs/cityscapes/mask_rcnn_r50_fpn_1x_cityscapes.py \
    checkpoints/mask_rcnn_r50_fpn_1x_cityscapes_20200227-afe51d5a.pth \
    8 \
    --format-only \
    --options "txtfile_prefix=./mask_rcnn_cityscapes_test_results"
1
2
3
4
5
6
7
8
9

错误

MMSeg错误：RuntimeError: Default process group has not been initialized

参考博客