使用 MMsegmentation 的2D语义分割网络测量物体的尺寸
从项目使用 zivid 3D 结构光相机采集数据，随后基于MMsegmentation框架中的 2D语义分割网络对2D图像进行分割，然后投影到3D数据中，分割出目标物体的3D信息。
搭建环境：
系统：Windows10
Mmcv-full版本为1.3.15
Torch1.8.0
Torchvision0.9
mmsegmentation 0.18.0
步骤1：
在anaconda中创建虚拟环境，python版本为3.6
步骤2：
安装CUDA、Pytorch
步骤3：
安装mmcv-full
pip install mmcv-full==1.3.15 -f https://download.openmmlab.com/mmcv/dist/cu102/torch1.6.0/index.html
测试：
在mmsegmentation目录下新建文件夹checkpoints
mkdir checkpoints
cd checkpoints
下载pspnet_r50-d8_512x1024_40k_cityscapes_20200605_003338-2966598c.pth
地址： https://github.com/open-mmlab/mmsegmentation/tree/master/configs/pspnet
cd ..\
cd demo
运行
python image_demo.py demo.png ..\configs\pspnet\pspnet_r50-d8_512x1024_40k_cityscapes.py ..\checkpoints\pspnet_r50-d8_512x1024_40k_cityscapes_20200605_003338-2966598c.pth
使用自己的数据集进行训练：
数据集准备：
使用labelme标注数据，然后将标注的json文件转换成png,生成两个文件夹:原图文件夹jpg和掩膜文件夹png (图像位深度为8，需要检查掩膜文件夹中的png文件是否图像位深为8，若有不为8的数据，需要进行转换, 转换后注意看标签的像素值是否发生变化）
配置相关参数：
相关的目录树
├─checkpoints(下载预训练权重,地址和测试时候的一样)
├─configs
│ ├─deeplabv3plus
│ └─deeplabv3plus_r50-d8_512x1024_40k_cityscapes_zhawa.py(复制deeplabv3plus_r50-d8_512x1024_40k_cityscapes改名)
│ └─_base_
│ ├─datasets
│ │ └─train_pratice.py(复制pascal_voc12.py改名)
│ ├─models
│ └─schedules
│ └─schedule_20k.py(修改)
├─data
│ ├─zhawa(新建)
│ ├─zhawa_result(网络训练好的权重保存地址)
│ ├─zhawa_test(测试图片分割预测结果保存位置，及距离测量结果保存)
│ ├─jpj(原图)
│ ├─png(mask)
│ ├─zdf
│ └─HD_zhawa(存放原始的zdf文件，其中的zdf文件名要与test.txt中的文件名一一对应)
│ └─splits(分离训练集、验证集和测试集)
│ ├─train.txt
│ ├─val.txt
│ └─test.txt
├─mmseg(修改以后将文件夹复制到 .../envs\mmdetection\Lib\site-packages，及环境中)
│ ├─datasets
│ ├─zhawa_voc.py(复制voc.py改名)
│ └─__init__.py(修改)
修改过的地方
deeplabv3plus_r50-d8_512x1024_40k_cityscapes_zhawa.py内容：
_base_ = [
'../_base_/models/deeplabv3plus_r50-d8.py',
'../_base_/datasets/train_pratice.py', '../_base_/default_runtime.py', //修改train_pratice.py
'../_base_/schedules/schedule_40k.py'
]
train_pratice.py的内容
# dataset settings
dataset_type = 'PascalVOCDataset_zhawa' //制作数据集时的类名字
data_root = '../data/zhawa/' //数据集的路径
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
crop_size = (768, 512)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='LoadAnnotations'),
dict(type='Resize', img_scale=(1024, 768), 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=(1024, 768),
# 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=2, //通过这个参数设置batch size
workers_per_gpu=2, //
train=dict(
type=dataset_type,
data_root=data_root,
img_dir='jpj',
ann_dir='png',
split = 'splits/train.txt',
pipeline=train_pipeline),
val=dict(
type=dataset_type,
data_root=data_root,
img_dir='jpj',
ann_dir='png',
split = 'splits/val.txt',
pipeline=test_pipeline),
test=dict(
type=dataset_type,
data_root=data_root,
img_dir='jpj',
ann_dir='png',
split = 'splits/test.txt',
pipeline=test_pipeline))
schedule_40k.py的内容
# 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=20000) //设置训练时网络迭代的epoch
checkpoint_config = dict(by_epoch=False, interval=200) //训练时每隔200epoch，网络保存一次权重
evaluation = dict(interval=200, metric='mIoU') //训练时每隔200epoch，网络使用验证集进行验证一次
train_pratice.py的内容
import os.path as osp
from .builder import DATASETS
from .custom import CustomDataset
@DATASETS.register_module()
class PascalVOCDataset_zhawa(CustomDataset): //将PascalVOCDataset 改为PascalVOCDataset_zhawa
"""train_db dataset.
In segmentation map annotation for Chase_db1, 0 stands for background,
which is included in 2 categories. ``reduce_zero_label`` is fixed to False.
The ``img_suffix`` is fixed to '.png' and ``seg_map_suffix`` is fixed to
'_1stHO.png'.
"""
CLASSES = ('background', 'tielian', 'ruanguan') //设置为自己数据集的标签名字，注意要加上background
PALETTE = [[128, 128, 128], [128, 0, 0], [0, 128, 0]] //给每一个标签设置一个颜色
def __init__(self, **kwargs):
super(PascalVOCDataset_zhawa, self).__init__( //将class的名称换为PascalVOCDataset_zhawa
img_suffix='.jpg',
seg_map_suffix='.png',
**kwargs)
assert osp.exists(self.img_dir)
__init__.py的内容
from .ade import ADE20KDataset
from .builder import DATASETS, PIPELINES, build_dataloader, build_dataset
from .chase_db1 import ChaseDB1Dataset
from .cityscapes import CityscapesDataset
from .custom import CustomDataset
from .dataset_wrappers import ConcatDataset, RepeatDataset
from .drive import DRIVEDataset
from .hrf import HRFDataset
from .pascal_context import PascalContextDataset
from .stare import STAREDataset
from .voc import PascalVOCDataset
from .train_db import TrainDBDataset
from .train_pratice import TrainPDataset
from .zhawa_voc import PascalVOCDataset_zhawa
__all__ = [
'CustomDataset', 'build_dataloader', 'ConcatDataset', 'RepeatDataset',
'DATASETS', 'build_dataset', 'PIPELINES', 'CityscapesDataset',
'PascalVOCDataset', 'ADE20KDataset', 'PascalContextDataset',
'ChaseDB1Dataset', 'DRIVEDataset', 'HRFDataset', 'STAREDataset',
'TrainDBDataset','TrainPDataset', 'PascalVOCDataset_zhawa'
]
生成split的代码
import numpy as np
filename_val='val.txt'
filename_train='train.txt'
train_list=[]
for i in range(1,1203):
train_list.append(i)
np.random.shuffle(train_list)
print(len(train_list))
with open(filename_train,'w') as file_object_train:
for n in range(0,1051):
numper_train=train_list[n]
s_train=str(numper_train)
file_object_train.write(s_train.zfill(4))
file_object_train.write("\n")
with open(filename_val,'w') as file_object_val:
for n in range(1051,1203):
numper_val=train_list[n]
s_val=str(numper_val)
file_object_val.write(s_val.zfill(4))
file_object_val.write("\n")
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent