实例分割SOLO & SOLOv2训练自建数据集(ubuntu20.04 + miniconda)

SOLO & SOLOv2训练数据集

接上一篇实例分割SOLO & SOLOv2环境配置(ubuntu20.04 + miniconda)
考虑到使用开源代码训练公开数据集缺少灵魂🤥，应该先使用开源代码训练自建数据集，然后就是使用改进的代码训练自建数据集🤤。 直接上流程😏

准备数据集

将数据集转成coco的数据格式，目录结构如下
不清楚怎么转coco数据格式的可以先移步搜索引擎，转好再回来

data
└── coco
    ├── annotations
    │   ├── instances_train2017.json
    │   └── instances_val2017.json
    └── images
        ├── test2017
        ├── train2017
        └── val2017

注册数据类别

在 mmdet/datasets/ 目录下创建自己的数据类别文件，例如 XDataset.py
增加代码如下

from .coco import CocoDataset
from .registry import DATASETS
 
#add new dataset
@DATASETS.register_module
class XDataset(CocoDataset):
    CLASSES = ['a', 'b', 'c']
# CLASSES里对应数据集的真实类别名

在 mmdet/datasets/__init__.py 文件中添加刚创建的数据格式 XDataset，修改后代码如下

from .builder import build_dataset
from .cityscapes import CityscapesDataset
from .coco import CocoDataset
from .custom import CustomDataset
from .dataset_wrappers import ConcatDataset, RepeatDataset
from .loader import DistributedGroupSampler, GroupSampler, build_dataloader
from .registry import DATASETS
from .voc import VOCDataset
from .wider_face import WIDERFaceDataset
from .xml_style import XMLDataset
from .XDataset import XDataset

__all__ = [
    'CustomDataset', 'XMLDataset', 'CocoDataset', 'VOCDataset',
    'CityscapesDataset', 'GroupSampler', 'DistributedGroupSampler',
    'build_dataloader', 'ConcatDataset', 'RepeatDataset', 'WIDERFaceDataset',
    'DATASETS', 'build_dataset', "XDataset"
]

修改配置文件

找到需要使用的模型配置文件，例如在 configs/solov2/xsolov2_light_448_r34_fpn_8gpu_3x.py 文件中更改检测类别，训练尺寸和数据类别，分别对应 num_classes ， img_scale 和 dataset_type ，修改后代码如下

# model settings
model = dict(
    type='SOLOv2',
    pretrained='torchvision://resnet34',
    backbone=dict(
        type='ResNet',
        depth=34,
        num_stages=4,
        out_indices=(0, 1, 2, 3), # C2, C3, C4, C5
        frozen_stages=1,
        style='pytorch'),
    neck=dict(
        type='FPN',
        in_channels=[64, 128, 256, 512],
        out_channels=256,
        start_level=0,
        num_outs=5),
    bbox_head=dict(
        type='SOLOv2Head',
        # a b c background
        num_classes=4, 
        in_channels=256,
        stacked_convs=2,
        seg_feat_channels=256,
        strides=[8, 8, 16, 32, 32],
        scale_ranges=((1, 56), (28, 112), (56, 224), (112, 448), (224, 896)),
        sigma=0.2,
        num_grids=[40, 36, 24, 16, 12],
        ins_out_channels=128,
        loss_ins=dict(
            type='DiceLoss',
            use_sigmoid=True,
            loss_weight=3.0),
        loss_cate=dict(
            type='FocalLoss',
            use_sigmoid=True,
            gamma=2.0,
            alpha=0.25,
            loss_weight=1.0)),
    mask_feat_head=dict(
            type='MaskFeatHead',
            in_channels=256,
            out_channels=128,
            start_level=0,
            end_level=3,
            num_classes=128,
            norm_cfg=dict(type='GN', num_groups=32, requires_grad=True)),
    )
# training and testing settings
train_cfg = dict()
test_cfg = dict(
    nms_pre=500,
    score_thr=0.1,
    mask_thr=0.5,
    update_thr=0.05,
    kernel='gaussian',  # gaussian/linear
    sigma=2.0,
    max_per_img=100)
# dataset settings
# dataset_type = 'CocoDataset'
dataset_type = 'XDataset'
data_root = 'data/coco/'
img_norm_cfg = dict(
    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
train_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(type='LoadAnnotations', with_bbox=True, with_mask=True),
    dict(type='Resize',
         # img_scale=[(768, 512), (768, 480), (768, 448),
         #           (768, 416), (768, 384), (768, 352)],
         img_scale=[(640, 480), (640, 420), (640, 400),
                    (640, 360), (640, 320), (640, 300)],
         multiscale_mode='value',
         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', 'gt_masks']),
]
test_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(
        type='MultiScaleFlipAug',
        # img_scale=(768, 448),
        img_scale=(640, 480),
        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']),
        ])
]
data = dict(
    imgs_per_gpu=2,
    workers_per_gpu=2,
    train=dict(
        type=dataset_type,
        ann_file=data_root + 'annotations/instances_train2017.json',
        img_prefix=data_root + 'images/train2017/',
        pipeline=train_pipeline),
    val=dict(
        type=dataset_type,
        ann_file=data_root + 'annotations/instances_val2017.json',
        img_prefix=data_root + 'images/val2017/',
        pipeline=test_pipeline),
    test=dict(
        type=dataset_type,
        ann_file=data_root + 'annotations/instances_val2017.json',
        img_prefix=data_root + 'images/val2017/',
        pipeline=test_pipeline))
# optimizer
optimizer = dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0001)
optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
# learning policy
lr_config = dict(
    policy='step',
    warmup='linear',
    warmup_iters=500,
    warmup_ratio=0.01,
    step=[27, 33])
checkpoint_config = dict(interval=1)
# yapf:disable
log_config = dict(
    interval=50,
    hooks=[
        dict(type='TextLoggerHook'),
        # dict(type='TensorboardLoggerHook')
    ])
# yapf:enable
# runtime settings
total_epochs = 300
device_ids = range(8)
dist_params = dict(backend='nccl')
log_level = 'INFO'
# 模型训练日志存放地址
work_dir = './work_dirs/solov2_light_release_r34_fpn_8gpu_3x'
load_from = None
resume_from = None
workflow = [('train', 1)]

训练数据集

打开终端，进入虚拟环境中运行代码

python tools/train.py configs/solov2/solov2_light_448_r34_fpn_8gpu_3x.py

测试权重

在 ./work_dirs/solov2_light_release_r34_fpn_8gpu_3x 文件夹下挑选希望使用的权重文件，更改 inference_demo.py 文件里 checkpoint_file 和对应的模型路径 config_file 以及测试图像读取路径 img

打开终端运行

python demo/inference_demo.py

注意！使用Pycharm等工具和终端直接运行检索的路径有区别，将出错路径改为绝对路径可以解决大部分路径问题